Channel Related Optimization of Wireless Communication Systems

Tufvesson, Fredrik

1998

Link to publication

Citation for published version (APA):

Tufvesson, F. (1998). Channel Related Optimization of Wireless Communication Systems Lund University

General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors
and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the
legal requirements associated with these rights.

• Users may download and print one copy of any publication from the public portal for the purpose of private study
or research.
• You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal
Take down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove
access to the work immediately and investigate your claim.

LUNDUNI
VERSI
TY

PO Box117
22100L und
+4646-2220000
Channel Related Optimization of
Wireless Communication Systems

Fredrik Tufvesson
Department of Applied Electronics
Lund University, Lund, Sweden

February 1998

Thesis for the degree of Licentiate in Engineering1

1
A Swedish university degree between M.Sc. and Ph.D.
© 1998 Fredrik Tufvesson

No. 3
ISSN 1402-8662
Printed in Sweden
Reprocentralen, Lund University
Lund 1998
Abstract

This thesis deals with different optimization problems in the design of wireless
communication systems. It is mainly directed to the design of systems based on
multicarrier techniques and orthogonal frequency division multiplex, OFDM, but
some of the problems apply to single carrier systems as well.
The influence of different pilot patterns is analyzed when pilot symbol assisted
modulation, PSAM, is used in OFDM systems. It is desirable to decrease the num-
ber of required pilot symbols and it is shown that the pilot pattern used plays a
major role to enable reliable channel estimates from a small amount of pilot sym-
bols. Rearrangement of the pilot pattern enables a reduction in the number of
needed pilot symbols up to a factor 10, still retaining the same bit error perfor-
mance.
The effect of the number of sub-channels used in an OFDM system is analyzed with
respect to resulting bit error rate. An analytical expression for the bit error rate on
Rayleigh fading channels when interchannel interference, ICI, caused by channel
changes during a symbol and energy loss due to the cyclic prefix are regarded. This
expression is used to optimize the number of sub-channels, and thereby the sub-
channel bandwidth (sub-channel spacing) in the system. It is argued that the system
can be optimized neglecting the effect of imperfect channel estimation and on a
worst case assumption for the Doppler frequency and signal to noise ratio.
The benefits of using pre-compensation (precoding) in wireless time division
duplex, TDD, systems are also investigated. The uplink channel estimate is used to
compensate the channel impact on the downlink symbols. This enables less complex
receiver structures in the mobile terminal since channel equalization is performed in
the base station. Three different methods where amplitude and/or phase are adjusted
are analyzed in terms of performance limits. Closed-form expressions for the QPSK
bit error rate are given assuming a fully known channel. It is shown that pre-com-
pensation is an attractive alternative to differential decoding. Phase-only compensa-
tion is preferred at low signal to noise ratios, while at high signal to noise ratios an
order of magnitude improvement in the bit error rate can be obtained by including
amplitude pre-compensation.
All the analyses and optimizations are general and can be applied to any OFDM
system.
Key words: Radio communication, multicarrier, orthogonal frequency division mul-
tiplex, channel estimation, pilot symbol assisted modulation, sub-channel spacing,
sub-channel bandwidth, interchannel interference, intersymbol interference, cyclic
prefix, Rayleigh fading, pre-compensation, precoding, time division duplex.

I
II
Preface

This licentiate thesis relies on the work I have carried out during the first three years
as a doctoral student at the Department of Applied Electronics. It comprises differ-
ent aspects of my research area, radio channel problems in conjunction with Ortho-
gonal Frequency Division Multiplex (OFDM) systems. The thesis consists of an
introduction and a background to OFDM and the channel problems, a discussion, a
summary and the papers listed below.
• Fredrik Tufvesson and Torleiv Maseng, ”Pilot Assisted Channel Estimation
for OFDM in Mobile Cellular Systems”, Proceedings of IEEE Vehicular
Technology Conference, Phoenix, U.S.A, pp. 1639-1643, May 4-7, 1997.
• Fredrik Tufvesson and Torleiv Maseng, ”Optimization of sub-channel band-
width for mobile OFDM systems”, in D. Everitt and M. Rumsewicz, editors,
Multiaccess, mobility and teletraffic – advances in wireless networks, pp.
103-114, Kluwer Academic Publishers, Dordrech, The Netherlands, 1998.
• Fredrik Tufvesson, Michael Faulkner and Torleiv Maseng, ”Theoretical anal-
ysis of pre-compensation for Rayleigh fading channels in QPSK TDD sys-
tems”, Series of technical reports, no. 3, ISSN 1402-8840, Department of
Applied Electronics, Lund University, Lund, Sweden, January 1998.
The last paper is not directed to OFDM systems only, but these systems are good
examples where the assumption of a flat slowly Rayleigh fading channel is met.
In addition to these papers I have written two conference papers that are not a part of
the thesis:
• Fredrik Tufvesson and Torleiv Maseng, ”On channel estimation strategies for
a cellular OFDM system”, Nordic Radio Seminar, Lund, Sweden, pp. 28-31,
August 27-28, 1996.
• Fredrik Tufvesson, ”Some channel estimation strategies for a cellular OFDM
system”, Proceedings of the RVK'96 RadioScience and Communication Con-
ference, Luleå, Sweden, pp. 161-164, June 3-6, 1996.

III
IV
Acknowledgement

I would like to thank all friends and colleagues at the Department of Applied Elec-
tronics for the help I have received during these first three years at the department.
Especially I am very grateful to some of the people in the radio communications
group, thank you Johan Hokfelt, Peter Malm, Ove Edfors and Anette Larsson for
valuable help and interesting discussions. This work is supervised by Professor Tor-
leiv Maseng. Without his help, sense and gut feeling for mobile communications it
could not have been done at all. I would also like to say thanks to Mike Faulkner,
who has inspired me with an enthusiasm that never dries up.
Finally, I am very grateful to Petra who has encouraged me all the time and to all of
my friends who have distracted me during the same period.
Financial support has been received from NUTEK (The Swedish National Board for
Industrial and Technical Development).

V
VI
Contents

1 Introduction 1

2 Background 3
2.1 The radio channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1.1 Reflections . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1.2 Transfer function . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.1.3 Coherence time and coherence bandwidth . . . . . . . . . . . 6
2.2 Single- or multicarrier systems? . . . . . . . . . . . . . . . . . . . . . 7
2.2.1 Examples of practical systems . . . . . . . . . . . . . . . . . . 8
2.3 OFDM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3.1 Cyclic prefix . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3.2 Sub-channel bandwidth . . . . . . . . . . . . . . . . . . . . .10
2.3.3 Transmission models . . . . . . . . . . . . . . . . . . . . . .11
2.3.4 Differential vs. coherent detection . . . . . . . . . . . . . . . .11
2.4 Channel estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.4.1 Pilot symbol assisted modulation . . . . . . . . . . . . . . . .14
2.4.2 Channel estimation in GSM . . . . . . . . . . . . . . . . . . .15
2.4.3 Channel estimation in DVB . . . . . . . . . . . . . . . . . . .16
2.4.4 How to design a pilot pattern . . . . . . . . . . . . . . . . . .17
2.5 Pre-compensation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.5.1 TDD systems . . . . . . . . . . . . . . . . . . . . . . . . . . .20
2.5.2 Channel compensation . . . . . . . . . . . . . . . . . . . . . .21
2.5.3 TDD in DECT . . . . . . . . . . . . . . . . . . . . . . . . . .22

A Some specifications of OFDM systems 23

VII
VIII

3 Paper summary 25

4 Discussion 27
4.1 New mobile communication systems . . . . . . . . . . . . . . . . . . 27
4.2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

Bibliography 29

Paper I
Pilot assisted channel estimation for OFDM in mobile cellular systems

Paper II
Optimization of sub-channel bandwidth for mobile OFDM systems

Paper III
Theoretical analysis of pre-compensation for Rayleigh fading channels in
QPSK TDD systems
Chapter 1

Introduction

The enormous increase in mobile telephone users makes the ether busy. New system
solutions are soon required to take care of all new subscribers and their demands.
The users do not just want their mobile phone for speech, but they also want the
possibility to send and receive fax, e-mail and data – without delay. The require-
ments of capacity and speed in the network entail that today’s systems have to be
developed further or replaced by new ones. This thesis deals with one of the trans-
mission techniques that can be used in these future mobile telephone systems, the
Orthogonal Frequency Division Multiplex technique or OFDM.
The current mobile telephone system in Europe, the GSM-system, was developed in
the late 80´s and then it was hard to foresee the enormous use of data communica-
tion that we have today. Few people had heard of the internet and the possibility of
sending e-mail and data files from a mobile terminal sounded like a fun but unneces-
sary feature of a phone. Besides, the expected number of subscribers was maybe a
tenth of the actual situation today, when about one third of the Swedes have their
own mobile phone. The GSM-system is mainly designed for speech and some draw-
backs have shown up, mostly in terms of capacity and speed. To get rid of some of
the problems the GSM-system will be further developed and probably some new
systems will be developed in parallel.
It is important that the new systems have high capacity, can provide high speed and
are flexible, i.e. that they can provide the data rates and services required by the
users both for speech and for high end applications such as video. High capacity is
achieved by transmitting low power, but then the signal is very sensitive to distur-
bances, which may be either noise or interference from other users. High speed can
be achieved by letting the time for one symbol, the symbol time, become very short.
But then problems with echoes due to reflections from buildings, mountains, cars
etc. arise. Finally, high flexibility is achieved by designing the system in such a way
that it can support different user requirements, but then it is important not to loose
efficiency in the transmission.
Obviously there always are a lot of considerations that need to be taken into account
in the design process. There are a lot of optimization problems that have to be
solved. Between the extremes there is hopefully a golden mean and this is what I try

1
2 1 Introduction

to find during my research, the golden mean for channel problems in conjunction
with one of the candidates for future generations of mobile telephone systems,
namely OFDM – Orthogonal Frequency Division Multiplex.
Chapter 2

Background

This chapter gives an introduction to OFDM, the radio channel and some of the
problems involved in mobile OFDM systems design. It is meant to serve as a back-
ground to the included papers and not as a complete description. First some basics
of digital communication are presented, then there is a description of the OFDM
technique and a detailed description of the channel estimation problem. Finally, a
technique called pre-compensation is presented, which is a method where the signal
to be sent is compensated in advance for channel impairments in order to get better
performance.

2.1 The radio channel

In digital communication systems the aim is to transmit digital bits from one point
to another. A channel is used in some way in order to transmit the information. The
channel may be the space, the sea or some sort of cable. Since this thesis is directed
towards mobile communications, the channel used in the following is generally the
air in which the radio wave propagates with reflections from surface objects. The
bits to be transmitted are mapped from ones and zeroes to waveforms suitable for
the radio channel, e.g. a high frequency cosine with phase shifts representing the
sent bits. The radio channel has some special properties which affect the design of
mobile communication systems, especially when it comes to the design of cellular
systems. Of particular interest are the reflections of the sent signal, the attenuation
of it, the changes of these parameters and the limited available bandwidth. Below
follows a description.

2.1.1 Reflections

Reflections are one of the main properties that make the design of a cellular system
challenging. They are caused by buildings, the ground, mountains, cars etc. which
are ”hit” by the radio wave, see figure 2.1.

3
4 2 Background

Figure 2.1: Reflections from a mobile terminal to a basestation.

The received signal can be described as a sum of all reflections, but since the radio
wave has different propagation ways there is a certain delay for each of the rays.
Mathematically the channel can be described by the impulse response [16]

Q
h ( τ, t ) = ∑ aq ( t ) ⋅ δ ( t – τq ( t ) ) , (2.1)
q=1

which is non-stationary due to movements of objects. In the above equation δ ( t )

denotes the Dirac delta-function and the index q represents the different echoes,
where q=1 means the first echo arriving to the receiver. An important channel
parameter is the maximum excess delay, which is a measure of the maximum differ-
ence in delay between the last and the first echo exceeding a specific level [23], i.e.
τQ ( t ) – τ1 ( t ) .

Reflections with the same delay reach the receiver at the same time and the sum of
these builds up a so-called tap. The tap coefficients, aq(t), describe the strength of
the reflection components (for each delay τ q ) and how they vary with time. The
result of the reflections is an amplitude change and a corresponding phase shift of
the sent signal. Since the delays and tap coefficients vary all the time the channel is
used to be characterized by some statistical model. One such model, used in the
GSM specification, is specified by the COST 207 group as a ”Typical Urban” chan-
nel [1]. In this model the relative power of the different taps is given, together with
the corresponding delay, see figure 2.2.
The statistics of the variations are also specified in the model. Often, the terminal
and some of the reflection points move and these movements result in a Doppler
shift. When the mobile terminal moves towards the receiver the received frequency
of the direct path is increased, but a reflex right behind the terminal will give a
decreased frequency. This means that the receiver at any one time will receive a
band of frequencies when reflections are present, even though only one frequency is
transmitted. This effect is called Doppler spread [22]. Clarke derived a channel
model [8] where it is assumed that the receiver receives reflections from all possible
directions. The resulting received spectrum has the form of a bowl centered around
the carrier frequency [8] [16], see figure 2.3, with edges at ± fD, the maximal Dopp-
ler frequency.
2 Background 5

average relative power (dB)

-2

-4

-6

-8

-10

-12
0 1 2 3 4 5
delay (µs)

Figure 2.2: Relative powers and delays in the COST 207 ”Typical Urban” model.
power spectrum density

-2 -1 0 1 2
frequency (f-fc)/fD

Figure 2.3: Received spectrum for Clarke’s channel model.

2.1.2 Transfer function

Sometimes it is desirable to analyze the influence of the channel in the frequency

domain instead of the time domain and then the transfer function is a valuable tool.
The transfer function, H(f,t), is simply the Fourier transform of the impulse response
and can be calculated as

Q
– j2πf τ q ( t )
H ( f , t ) = F { h ( τ, t ) } = ∑ aq ( t ) ⋅ e . (2.2)
q=1

It describes the attenuation and phase shift of e.g. the channel for each of the fre-
quency components in the transmitted signal.
6 2 Background

2.1.3 Coherence time and coherence bandwidth

The autocorrelation function, ρ T ( ∆t ) , of a variable is an important measure of how a

certain variable is dependent on itself from one time instant to another. For station-
ary processes it can be calculated as the expectation of the value of a variable multi-
plied by the value of the same variable at a time ∆t later, i.e.

ρ T ( ∆t ) = E [ x ( t )x ( t + ∆t ) ] . (2.3)

For Clarke’s channel model the autocorrelation function of the received envelope is
a zeroth order Bessel function of the first kind, which is dependent on the Doppler
frequency and the time delay [16],

ρ T ( ∆t ) = J 0 ( 2π f D∆t ) . (2.4)

The autocorrelation function is important because it describes how samples of the

channel are coupled to each other and this information is often used in channel esti-
mation filters. There is also an autorcorrelation function in frequency which
describes the coupling between the transfer function values for a certain frequency
separation. The frequency correlation is dependent on the carrier frequency used,
the power delay profile of the channel and mean power of the different taps. In order
to describe the time and frequency range over where the correlation is high one can
use the coherence time and coherence bandwidth. The coherence time is defined as
the time span where the correlation exceeds a specific limit, e.g. ρ T ( ∆t ) =0.5,
whereas the coherence bandwidth is defined as the bandwidth where the correlation
between the channel transfer functions ρ F ( ∆f ) exceeds the same value. The coher-
ence time, Tc, and the coherence bandwidth, Bc, can be estimated as [22]

1 1
T c ≈ ------ and B c ≈ ------------------------------ . (2.5)
fD τQ ( t ) – τ1 ( t )

The coherence time is inversely proportional to the Doppler frequency and can be
interpreted as ”the time over which the channel is constant”. The coherence band-
width is inversely proportional to the excess delay and can in the same manner be
seen as a measure of ”the bandwidth over where the channel is flat”. That means
that the coherence time is short when a mobile terminal moves fast and the coher-
ence bandwidth is small when long delayed reflections are present, e.g. in hilly ter-
rains. An example of the correlation function for the ”Typical Urban” channel [1] is
shown in figure figure 2.4.
2 Background 7

0.8
abs(correlation)
0.6

0.4

0.2

500
0.6
0.4
0 0.2
0
-0.2
-500 -0.4
-0.6
freq. dist. (kHz) time dist. fD*t

Figure 2.4: Time- and frequency correlation for a ”Typical Urban” channel.

2.2 Single- or multicarrier systems?

Reflections cause problems in communication systems since the signals from differ-
ent echoes arrive at different time instants to the receiver. This means that the
received signal is a sum of the signal sent at different time instants and so-called
inter symbol interference, ISI, arises [22]. When the symbol time is short compared
to the excess delay then the ISI spans over several symbols and this effect has to be
suppressed by the receiver. Normally this can be accomplished by an equalizer, e.g.
a Viterbi equalizer [22], but when the ISI spans over many symbols the equalization
procedure gets quite complicated. Today there is a growing interest for high speed
mobile communication, and when the symbol rate is high, the symbol time is short
and there is a risk that the ISI becomes severe. One way to get around this problem
is to divide the data stream into several sub-streams and transmit each of them on its
own frequency. Then the symbol time on each carrier frequency gets long even
though the overall bitrate is high, and the problem with ISI originating from several
symbols can be solved. However new problems arise as the duration of each symbol
becomes long compared to the coherence time. The channel variations get fast com-
pared to the symbol duration, which cause problems for the receiver and degrade the
performance. Problems also arise when a transmitter has to transmit several signals
at the same time. One of the reasons that single carrier systems are more popular
than multicarrier systems is that the envelope of the sent signal may be constant.
This is not the case in multicarrier modulation and therefore it is hard to design a
power amplifier with high efficiency.
In order to get a spectrum efficient multicarrier system the sub-carriers have to be
placed with minimum frequency distance between them but with enough distance
such that they do not interfere with each other. This problem was solved in the late
60´s [7] with overlapping spectra between the carriers, but with maintained orthogo-
nality between them and the first ”modern” OFDM system was invented.
8 2 Background

2.2.1 Examples of practical systems

In the following sections we will use some existing systems, GSM, DECT, DAB
and DVB, in order to discuss and relate parameters to what is used in reality.
GSM [11] is the pan European system for mobile telephony specified in 1990. It is
mainly designed for speech, but offers some data services as well. GSM is a TDMA
FDMA based system, often referred to as a second generation system, with eight
users per channel and a channel data rate of 271 kbit/s. The system was originally
designed for the 900 MHz band, but is now also used in the 1800 MHz band.
DECT [11] is the European digital cordless telephone standard. It is designed to
offer a broad range of services for short range high density telecommunications.
The system relies on a TDMA FDMA TDD structure and is not as ”centralized” as
GSM. The base stations are designed to work more independently of each other. The
standard was finalized in 1992 and the system works in the 2 GHZ band.
DAB refers to the digital audio broadcasting system in Europe [3], see appendix A
for details. It was standardized in 1995 and is a multicarrier broadcast system for
”near CD” quality in mobile, portable and fixed receivers. Besides the audio ser-
vices the standard also supports transmission of multimedia signals. All the services
are multiplexed into one data stream before transmission and a single frequency net-
work is possible for transmission. This means that the same frequency can be used
for simultaneous transmission of the same signal from many transmitter sites.
DVB is the new European standard for digital video broadcasting [2]. It is also a
multicarrier system, but designed for higher data rates than the DAB system since it
is intended for video transmission. The system supports single frequency networks
as well and is meant for transmission on the same frequencies as the analog TV sys-
tem.

2.3 OFDM

This section presents the basics of OFDM needed to understand the problems lying
behind the included papers. It is not meant to give a complete overview of the tech-
nique, more exhaustive presentations of the OFDM concept can be found in e.g.
[12], [4] and [9].
As mentioned before, the information can in OFDM systems be seen as transmitted
on several sub-carriers. In principle an OFDM transmitter works in the following
way [5]. The symbols to be transmitted are fed to a serial to parallel converter and
then an inverse fast Fourier transform, IFFT, is performed. The samples from the
IFFT are then converted to serial form, extended by a cyclic prefix (explained later),
mixed to an appropriate frequency and transmitted over the radio channel, see figure
2.5. It is the IFFT that transforms the signals such that the samples from the parallel
to serial converter constitute a signal where the data can be seen as transmitted on
several sub-carriers. All of the values from one IFFT constitute samples from one
OFDM symbol. Therefore the symbol time for the OFDM symbol is M times longer
than the symbol time for the data symbols [4], where M is the number of points in
2 Background 9

…
data S/P IFFT P/S

Figure 2.5: Block description of an OFDM transmitter

the IFFT. In this way it is possible to extend the OFDM symbol time, and get it
much longer than the maximum excess delay, thereby making the echoes affect only
the first part of the next symbol, but not more.

2.3.1 Cyclic prefix

Even though the symbol time is much longer than the maximum excess delay, there
is still some intersymbol interference, ISI. The so-called cyclic prefix, which was
introduced in the early 80’s in [18], may be used to get around this problem. The
receiver can handle echoes within one OFDM symbol since these only result in a
phase and amplitude change, but it can not handle echoes between the symbols. In
the following it is described how the idea of the cyclic prefix can be used to handle
multipath delay such that the demodulated signal is free of both intersymbol inter-
ference and interchannel interference, ICI.
The solution to the problem is to make the echoes affect only one symbol at a time
by extending the symbol time. The last samples from the IFFT are then copied and
transmitted before the first IFFT samples. At the receiver side the receiver discards
the (just added) first samples of the received symbol, and the original first samples
are now affected only by echoes from its own OFDM symbol, and not from previ-
ous symbols, see figure 2.6. Mathematically speaking: The linear convolution per-
formed by the channel is transferred into a cyclic convolution [9], which after the
FFT acts like a scalar multiplication by the channel transfer function.
one symbol
ISI

copy copy copy

CP no ISI CP CP

Figure 2.6: Insertion of the cyclic prefix, CP, means that the ISI can be removed by the
receiver without loss of orthogonality between the symbols.
10 2 Background

No ISI arises as long as the length of the cyclic prefix exceeds the maximum excess
delay, i.e. the difference in delay between the last and first echo. However, the use of
the cyclic prefix has some drawbacks. Naturally, the total symbol time including the
cyclic prefix has to be longer than the maximum excess delay, which in turn limits
the signalling rate on each sub-channel. There is also a waste of power since the
transmitter discards the energy spent on the cyclic prefix. In the early OFDM sys-
tems [31] a guard space, a period of silence, was used between the symbols instead
of the cyclic prefix. Then ISI is avoided but the sub-channels are not entirely orthog-
onal and some ICI arises. Anyhow the guard space or transmission of the cyclic pre-
fix takes some extra time and the bitrate is degraded. The last problem can be solved
by faster signalling, but then more bandwidth is occupied instead. So, one has to
choose the ”golden mean” between having enough cyclic prefix in order to avoid or
suppress ISI, but short enough not to waste too much energy or bandwidth.

2.3.2 Sub-channel bandwidth

In OFDM systems the total bandwidth can be seen as divided into sub-channels and
therefore we can associate a specific sub-channel bandwidth, or sub-channel spac-
ing, with them. In the following we use the term sub-channel bandwidth to denote
the distance in frequency between the different sub-channels even though the actual
bandwidth of each sub-channel is much larger. Figure 2.7 shows a symbolic plot of
four sub-carriers in an OFDM system.

sub-channel
bandwidth
signal spectrum

frequency

Figure 2.7: Symbolic representation of four sub-channels in an OFDM system.

The bandwidth is dependent on the data rate on each of the sub-channels. Many sub-
channels means that the data rate on them is low and then the corresponding band-
width becomes low, and vice versa. The advantage of having many sub-channels is
that the duration of the cyclic prefix becomes short compared to the symbol time,
which reduces the waste of bandwidth and energy. But, the orthogonality between
the channels relies on the assumption that the channel characteristics remains con-
stant during a symbol interval. The sub-channels start to disturb each other when
changes of the echo-pattern during a symbol interval becomes perceptible [26] and
interchannel interference arises. The symbol time gets long when many sub-carriers
are used and the channel changes caused by e.g. movements of the mobile terminal,
2 Background 11

become evident. Therefore the choice of sub-channel bandwidth is not obvious. On

the one hand we want to have several sub-channels and long symbols in order to
keep the duration of the cyclic prefix small. On the other hand we want to have few
sub-channels and short symbols in order to avoid channel changes during a symbol
and maintain orthogonality between the sub-channels. The second paper included in
this thesis [29] treats this problem and how to find the trade-off between the
extremes.

2.3.3 Transmission models

An OFDM system is free of both ISI and ICI if the coherence bandwidth is much
greater than the sub-channel bandwidth, the channel changes are negligible during a
symbol interval and the length of the cyclic prefix is long enough to exceed the max-
imum excess delay [9]. Since the sub-channels are not influenced by each other it is
possible to define a model for the OFDM system where the signal is transmitted on
separate Gaussian channels [9]. The sent data symbols are phase shifted and attenu-
ated by the transfer function, Hk,m, of the sub-channel, m, at time, k. Finally the
symbols are disturbed by white Gaussian noise, nk,m, see figure 2.8.

Hk,0 nk,0
xk,0 rk,0

Hk,1 nk,1
xk,1 rk,1
…

Hk,M-1 nk,M-1
xk,M-1 rk,M-1

Figure 2.8: OFDM system model when ISI and ICI are avoided.

Note that no complex equalizer is needed in the receiver, just a phase shift and
maybe an amplitude compensation for each carrier. This is the one of the main rea-
sons why OFDM is attractive for mobile communication.
The transfer functions are correlated with each other as described in section 2.1.3. If
the transfer function of a specific channel at a certain time is bad, i.e. has a low
amplitude, then the adjacent transfer functions are quite likely to be bad too, and
vice versa. Decreasing the sub-channel bandwidth increases the frequency correla-
tion and decreases the time correlation of the different transfer functions.

2.3.4 Differential vs. coherent detection

Either differential or coherent detection can be used when demodulating the

received signal [22]. Differential detection of PSK-signals means that the receiver
makes its decisions based on the phase difference between two successive symbols
12 2 Background

instead of the absolute phase. The phase difference can be achieved by multiplying
the received symbol value by the complex conjugate of the previous or adjacent one,
see figure 2.9.

Hk,m nk,m
xk,m yk,m

x* T

Hk,m-1 nk,m-1
xk,m-1 yk,m-1

Hk,m nk,m x*
xk,m yk,m

Figure 2.9: Calculation of decision variables for differential detection.

Differential detection in time (top) and in frequency (bottom).

The resulting decision variable is in case of differential detection the phase of [22]
either

y k, m = ( x k, m H k, m + n k, m ) ( x k – 1, m H k – 1, m + n k – 1, m )∗ (2.6)

or

y k, m = ( x k, m H k, m + n k, m ) ( x k, m – 1 H k, m – 1 + n k, m – 1 )∗ (2.7)

depending on whether differential signalling in time or in frequency is used. Nor-

mally the one is chosen where the channel correlation is the largest since the
changes should be caused by data and not by the channel. In the ideal case the trans-
fer function is constant but since we use the difference between two disturbed values
we get two noise contributions. If the transfer functions differ for the two symbols,
we get an additional deviation from the ideal value.
Coherent detection is based on one received value at a time and relies on the knowl-
edge of a reference phase and amplitude for the symbols. The received symbol val-
ues are compared with the reference and then decisions are made. The reference
phase and reference amplitude are obtained by estimating the influence of the chan-
nel on the sent symbols, and the goal is to undo this influence. The decision variable
for coherent detection is

y k, m = ( x k, m H k, m + n k, m ) ⁄ Ĥ k, m (2.8)
2 Background 13

where Ĥ k, m is an estimate of the transfer function at time k of channel m. For PSK

systems only the phase is interesting so that equation (2.8) may be changed to

y k, m = ( x k, m H k, m + n k, m )Ĥ∗ k, m , (2.9)

which is shown in figure 2.10.

Hk,m nk,m Ĥ k, m∗
xk,m yk,m

Figure 2.10: Decision variable for coherent PSK detection.

If the channel is known then Ĥ k, m = H k, m and compared to equation (2.6) we have

only one noise contribution in this case. This means that if it is possible to know the
channel impact in some way, then we can accept around two times (3 dB) more
noise for the same performance. One may argue that if perfect coherent detection
could be performed, then around 50% of the power could be spent on pilot symbols
for channel estimation and synchronization. The question is how to get this channel
information without too much loss. If the channel estimate is bad, we may get a per-
formance which is worse, where the 3 dB gain may be turned into a loss instead.
However, it turns out that only a few percent of pilot symbols is sufficient to enable
reliable channel estimates [20] which can be used for coherent detection.
The choice between coherent or differential detection is also a question of receiver
complexity. In some cases it is desirable to have cheap receivers and then it is better
to use differential detection to get rid of the channel estimation problem. In [24] it is
argued that the losses due to channel estimation is in the same region as the 3 dB
loss for differential detection and that there is no reason for using coherent detection
in a mobile communication system. This depends on how channel estimation is per-
formed and if the channel estimation is optimized for the actual propagation condi-
tions it is shown in [6] that a coherent system has better performance.

2.4 Channel estimation

As seen in the last section, the receiver has to know the transfer functions of the dif-
ferent sub-channels and time instants in order to enable coherent demodulation.
These transfer functions can be estimated by transmitting known signals, so-called
pilot symbols or pilot tones, and comparing these with the received signals. Since
both the sent phases and amplitudes are known, it is easy to estimate the channel
influence on the channels where the pilot symbols are sent. Then the influence on
the channels situated in between, i.e. the data channels, can be achieved by interpo-
lation. The pilot symbols transmit no data and therefore it is desirable to use as few
of them as possible. They occupy bandwidth which rather could be used for data
transmission and, if the total energy is to be kept constant, they lead to a power loss
for the data symbols. However, the number of pilot symbols can not be decreased
arbitrarily since a certain amount is needed to enable reliable channel estimates.
14 2 Background

It is also possible to use the data symbols as virtual pilot symbols [10]. The received
signal is demodulated and if a decision is correct the corresponding symbol can be
used as a known pilot symbol. The technique is called decision directed channel
estimation and means that there is no need for extra known symbols to be sent. This
of course increases the overall capacity compared to a pilot symbol based system if
the decisions are correct. But the receiver does not know whether a decision is cor-
rect or not and the result is longer error bursts after rapid fades and slower acquisi-
tion.

2.4.1 Pilot symbol assisted modulation

Pilot symbol assisted modulation, PSAM, is a channel estimation technique where

pilot symbols are inserted periodically into the data stream [6]. No special tones are
reserved for pilot information and the characteristics of the sent signal remain the
same, independent of whether data symbols or pilot symbols are sent. Cavers made
an exhaustive analysis in [6] for the single-carrier case, and the extension to multi-
carrier systems is straightforward. In multicarrier systems we have the opportunity
to interpolate the received channel information both in the time direction and in the
frequency direction. This means that it is possible to use less pilot symbols for chan-
nel estimation, but nevertheless get reliable estimates. One can think of a time fre-
quency grid where the impact of the radio channel has to be estimated. In the single
carrier case this grid is very narrow and the pilot symbols can be put in a row along
the time direction only. But, in the multicarrier case the time frequency grid is
broader and the freedom to place pilot symbols is much greater, see figure 2.11.
f

single carrier
f

multicarrier
Data Pilot

Figure 2.11: In multicarrier systems the pilot symbols can be spread in both the
time and the frequency direction which decrease the number of required pilot
symbols.
2 Background 15

Paper one [28] addresses the problem of where and when to insert pilot symbols in
the time-frequency grid. Different pilot patterns are analyzed and we examine the
difference between single carrier and multicarrier systems. It is shown that it is pos-
sible to reduce the number of required pilot symbols a lot, up to ten times, by using
a multicarrier system where the pilot symbols are placed in a pattern where both the
correlation in time and in frequency can be utilized by the channel estimator.
The channel information derived from the pilot symbols has to be interpolated in
some form in order to get the channel estimate for the data symbols as well. The
optimal filter, in the sense that the mean square error is minimized, is the 2-dimen-
sional Wiener filter [15]. The Wiener filter relies on a knowledge of the actual corre-
lation between the data symbol of interest and all of the pilot symbols and between
all the pilot symbols themselves. The channel estimate for a point is a weighted sum
of all received pilot values, where pilot information in the vicinity (in terms of high-
est correlation) of the point have high weights. Since all pilot symbols contribute to
all channel estimates in a block, large matrix operations are needed and therefore
the filter becomes computationally heavy. In practical systems, some kind of simpli-
fications have to be made, see e.g. [9] and [27] for detailed descriptions. One
straightforward method is to use two 1-dimensional Wiener filters and interpolate in
one direction at a time, e.g. first in the time direction and then in the frequency
direction. The performance is similar to that of 2-dimensional filtering [15] and the
complexity is reduced since no huge matrix operations has to be made. Another pos-
sible alternative is to use only the pilot symbols in the neighborhood for the estimate
[9]. These are the most important ones since they have the highest correlation and
pilot symbols far away from the actual point do not anyhow significantly contribute
to the estimate.

2.4.2 Channel estimation in GSM

The GSM system is designed as a single carrier system. Therefore channel estima-
tion in GSM is a little bit different from that in OFDM systems. The channel estima-
tion procedure relies on the reception of known bits, which are called a training
sequence, but in GSM it is the time domain properties of the channel that are of
main interest [24]. The training sequence is located in the middle of each GSM
burst, see figure 2.12. Each burst consists of 168 symbols and 26 of these (≈18%)
constitute the training sequence which is used for estimation and synchronization. A
GSM burst has a duration of 546 µs and a good approximation to make a simple
receiver is to assume that the channel remains constant during this time. This is not
a necessary assumption but works well since time correlation of the channel is
rather high within one burst. The channel correlation is approximately 0.97 from the
middle of the burst to the end when the terminal is moving at a speed of 250 km/h,
the worst case speed in the GSM channel models [1].
The receiver performance for a given channel is not specified in the standard, but a
maximum excess delay of 16 µs is specified for the equalizer test channel model [1].
The type of equalizer to be used is neither given, however a Viterbi equalizer is
commonly used [34]. Since the symbol time in GSM is 3.7 µs, the channel estimator
and equalizer have to mitigate ISI which may span over 5 symbols. The 8 training
16 2 Background

GSM frame
4.61 ms

user: 0 1 2 3 4 5 6 7 0

GSM burst

flag

flag
tail

tail
data training data
bits: 3 57 1 26 1 57 3 8,25

546 µs 30µs

Figure 2.12: The structure of a GSM frame and a normal GSM burst [24].

sequences used for channel estimation consist of 26 symbols with good autocorrela-
tion properties, i.e. an autocorrelation with a sharp peak for a certain delay but close
to zero elsewhere.
The channel estimates are achieved by correlating the received training sequences
with the known one, which is stored in the receiver. Since the autocorrelation func-
tion has a sharp peak, an estimate of the impulse response of the channel can be
achieved from this correlation. This impulse response estimate is then fed to a
Viterbi equalizer, which makes a decision in favor of the most probable sent
sequence when the received one and the channel impulse response is known.

2.4.3 Channel estimation in DVB

OFDM and coherent demodulation is used in the Digital Video Broadcasting [2],
DVB, system. It is mainly meant for stationary receivers, so the problems with
Doppler broadening and an altering channel is not as big as in a mobile communica-
tion system, but it is anyhow interesting to study the channel estimation structure.
The pilot symbols are used for several purposes, e.g. frame synchronization, fre-
quency synchronization, time synchronization, channel estimation and phase noise
tracking. In order to assist all these fields of applications the pilot symbols are
located both at continuous pilot channels and as scattered pilots. There are in total
6817 sub-channels (in the 8k mode) and 177 of them are dedicated as pilot chan-
nels. The scattered pilots are located at every 12th sub-channel with a frequency
shift of three sub-channels every new symbol, see figure 2.13.
The total pilot density is approximately 10%, which should be large enough to
enable reliable channel estimates. For channel estimation only the pilot density may
seem to be high, but the pilot symbols are used for other purposes too and robust
detection is important when modulation schemes sensitive to estimation errors are
used, e.g. 64-QAM.
2 Background 17

Data
Pilot

Figure 2.13: Example of pilot location in DVB.

2.4.4 How to design a pilot pattern

The most important parameters for the design of a pilot pattern are the expected
maximum speed, which determines the minimum coherence time, and the maxi-
mum excess delay, which determines the minimum coherence bandwidth. The pilot
symbols have to be placed close enough in order to be able to follow the time and
frequency variations of the transfer function, but far enough not to increase the over-
head too much. The lower limit for the pilot density is determined by the Nyquist
sampling theorem if one wish to be able to follow all the variations, but in practice
one has to sample the fading process more often, i.e. insert pilot symbols, in order to
get reliable channel estimates. In [15] it is suggested as a ”rule of thumb” to use
twice as many pilot symbols in time respectively frequency as stated by the sam-
pling theorem. A suitable choice of the pilot spacing in time, NT, and in frequency,
NF, is therefore

1 1
N T ≈ --- ⋅ --------------------- (2.10)
2 2 f D T sub

and

1 1
N F ≈ --- ⋅ --------------------- (2.11)
2 2∆f τ max

where ∆f is the sub-carrier bandwidth, Tsub is the symbol time and τ max the maxi-
mum excess delay of the channel. The influence of the pilot density is also investi-
gated in [20] with similar conclusions. If a low complex channel estimator is used,
e.g. two 1-dimensional estimators instead of one 2-dimensional estimator, the sys-
tem is more sensitive to the choice of the pilot density, but the ”twice the Nyquist
frequency rule” seems to work well even in this case [20]. The relation between the
pilot spacing in time and in frequency is important to minimize the pilot density. It
is advantageous to have the same ”uncertainty” in both the time direction and in the
18 2 Background

frequency direction, to have a ”balanced” [15] pilot pattern. This is accomplished by

the design rules above and can be viewed as having the same distance between the
pilot symbols in both directions, when normalized by the coherence bandwidth and
the coherence time. If we define the coherence time as 1/fD and the coherence band-
width as the inverse of the excess delay, 1/ ( τ Q ( t ) – τ 1 ( t ) ) , then suitable pilot spacings
are one fourth of the coherence time and coherence bandwidth respectively.
If the channel correlation functions are not fully known it is important to measure
the ”edge symbols”, i.e. the outer channels in frequency and, if a block structure is
used, the first and last symbols in each block. Then filtering and smoothing can be
used and prediction can be avoided. In practical situations the channel characteris-
tics is often not known since it is preferable to avoid requirements of e.g. measuring
the Doppler frequency. For simplicity and robustness of the channel estimator it is
often suggested to adjust it according to the worst case scenario [14] for the Doppler
frequency and excess delay, i.e. to the minimum correlation in time and frequency.
In order to avoid prediction pilot symbols are therefore suggested to be placed at the
edge channels and then sufficiently many pilot symbols are uniformly placed
between them in order not to exceed the pilot spacings given by the expressions
(2.10) and (2.11).
The guidelines above are derived for a Rayleigh channel. If the channel sometimes
is Ricean distributed and sometimes Rayleigh distributed the pilot pattern has to be
designed for the worst case, i.e. the Rayleigh one, since the channel fluctuations are
faster there. In the general case, one has to insert pilot symbols often enough to be
able to follow the variations in time and frequency. What ”often enough” means is
determined by the Doppler spectra and the power delay profile for the whole sys-
tem, i.e. including any hardware impairments such as oscillator drift and phase
noise. The importance of the latter should not be underestimated when it comes to
the design of new systems working at frequencies of several GHz.
The effect of different pilot densities is viewed in figure 2.14. For simplicity a single
carrier system is assumed where every tenth symbol is a pilot symbol. If the symbol
time is constant we may vary the doppler frequency in the system in order to study
the effect of different pilot densities, or sampling rates compared to the Doppler fre-
quency. Figure 2.14 shows the resulting mean bit error rate for 101 QPSK symbols
when the Doppler frequency, which is assumed to be known by the receiver, is var-
ied. The ”Typical Urban” channel [1] is estimated by a Wiener filter and the result-
ing bit error rate is calculated using the method in the appendix of paper II.
For low signal to noise ratios the bit error rate is determined by the noise in the
channel and the channel estimates are sufficiently good even for rater low pilot den-
sities. But, as the signal to noise ratio increases, the bit error rate reaches an error
floor, which is determined by the pilot density and the channel estimation errors.
The crossing between the noise limited region and the channel estimate limited
region is where the variance of the noise and the variance of the estimation error are
equal. The estimate limited region can be avoided if the pilot density is high enough.
For practical signal to noise ratios the expressions (2.10) and (2.11) seem to work
well.
The bit error rate has similar behavior in case of OFDM with 2-dimensional filter-
ing. The mean bit error rate for 30 sub-channels and 30 OFDM symbols is presented
in figure 2.15. The levels of the estimate limited regions are somewhat lower com-
2 Background 19

0
10

1.0
-2
10
BER
1.2
-4
10
1.4
1.6
1.8
-6
2.0
10 coh.
0 10 20 30 40 50
Eb/No (dB)

Figure 2.14: Mean bit error rate for different oversampling factors in a single
carrier system. 1.0 corresponds to the Nyquist sampling rate.

0
10

0.4
-2
10
BER

0.6

10
-4 0.8
1.0
1.2
1.6, 2.0
-6 coherent
10
0 10 20 30 40 50
Eb/No (dB)

Figure 2.15: Mean bit error rate for different oversampling factors in a
multicarrier system. 1.0 corresponds to the Nyquist sampling rate.

pared to the single carrier case due to the suppression of the peaks in two dimen-
sions. This indicates the benefit of using some kind of 2-dimensional channel
estimation, where filtering in the time direction as well as in the frequency direction
is used.

2.5 Pre-compensation

Pre-compensation is a technique where the signal to be sent is compensated in

advance for e.g. channel impairments in order to achieve better performance. The
technique requires reliable estimates of the actual transfer function or impulse
response. These estimates may for example be obtained by using pilot symbols in
the same way as described in section 2.4. Pre-compensation is rather straightfor-
ward if a so-called TDD structure is used in the system, since the channel estimate
for the incoming signal can be used for the outgoing signal as well.
20 2 Background

2.5.1 TDD systems

Time division duplex, TDD, is a duplex technique where the same frequency is used
both by the base station and the mobile terminal, but at different time instants. Think
for example of a TDMA system with two time slots. If the user data rate is much
smaller than the channel data rate, then one single channel may appear as a full con-
tinuous duplex connection to the user [23]. In TDD systems the possibility of mak-
ing a prediction of the channel influence is rather straightforward. The used radio
channel is often assumed to be reciprocal, see e.g. [18] [33], since the echo pattern
and thereby the channel impulse response is the same both for uplink transmission,
the transmission block from the mobile terminal to the base station, and downlink
transmission, the block from the base station to the terminal. If the blocklength is
short compared to the speed of change of the channel, then a channel estimate of the
uplink block is quite likely to be valid for an adjacent downlink block too. This fact
can be used to change the signal to be sent in a way such that better signal to noise
ratio is achieved at the receiver or such that a more simple receiver can be used. Fig-
ure 2.16 shows the principle of the technique, often referred to as pre-coding or pre-
compensation, used in an OFDM-TDD system.

estimate channel estimate channel

f pre-compensate pre-compensate
uplink downlink uplink

delay blocklength Data Pilot

Figure 2.16: Block structure of a pre-compensated OFDM-TDD system.

The TDD structure can be used for different of pre-compensation methods, for
example to control the output power [19] to achieve better performance or to modify
the sent signal in order to suppress ISI [18]. The technique can also be used to
implement a rather simple antenna diversity system at the base station which results
in a diversity gain for both the base station and the mobile terminal [13]. Think for
example of a system where two antennas are used at the base station. They are both
receiving and measuring the incoming signal on the uplink. On the downlink the
best antenna is used for transmission, thereby avoiding possible fading dips at the
other antenna.
Between the uplink block and the downlink block there is often a delay [11], which
is used to secure non overlapping transmission from the two transmitters. In a pre-
compensated system the base station has to perform all the calculations and com-
pensations within this delay so there is a need for fast processing in the base station.
2 Background 21

2.5.2 Channel compensation

Channel estimation is a rather complex operation to perform, as seen in section 2.4.

Every terminal has to make channel estimates in order to use coherent detection. By
pre-compensation it is possible to move the task of channel estimation to the base
station, but still achieve the performance benefits of coherent detection in the mobile
terminal. Then cheap, less complex, mobile receivers can be used. The channel esti-
mators at the base stations are used both for uplink and downlink equalization.
Complexity and cost is not as important there since the equipment is shared among
many users and connected to the main power system. The technique relies, as men-
tioned before, on a knowledge at the base station of the expected channel influence
on the signal to be transmitted. The basestation estimates the channel influence, e.g.
the transfer functions or the impulse response, on the downlink symbols and com-
pensates the downlink symbols for this in advance, see figure 2.17.

Hk nk
xk xk H k + nk

Hk + l xk + l Ĥ k
-----------
- xk + l + nk + l ----------
Ĥ k Ĥ k

nk+l Hk+l

Figure 2.17: The principle of channel pre-compensation in an OFDM system.

If the compensation is correct, the influence of the channel is imperceptible to the

receiver. Paper three [30] is an investigation of the theoretical performance of three
different pre-compensation strategies when the channel is known. However, the
channel transfer functions are likely to change as time passes. There is a maximum
blocklength to be used, which is determined by the coherence time, before the per-
formance has degraded too much.
The technique has shown to be promising for mitigating multipath distortion [18] of
the radio signal. The latter is particularly interesting in e.g. CDMA systems where a
complex so-called RAKE receiver often is necessary to mitigate the multipath
spread. By pre-compensation the multipath problem can be taken care of at the
basestation where one can afford more advanced and complex receivers.
A pre-compensated system works like when two persons are playing outdoor ping-
pong. In order to hit the table the players have to compensate for the wind and its
force on the ball. They can observe the influence of the wind when the ball comes
towards them and hopefully take the appropriate action before they have to hit the
ball again.
22 2 Background

2.5.3 TDD in DECT

DECT, Digital European Cordless System, is a standard for cordless telephony,

mainly devoted to private systems within a home or office. It may however be used
in metropolitan areas too or in conjunction with an area covering system such as
GSM for ”normal” mobile services [22]. The standard specifies a combination of
FDMA, TDMA and TDD. A DECT frame consists of 24 time slots, 12 of them are
used for uplink transmission and 12 for downlink, see figure 2.18.
DECT frame
uplink downlink
0 1 2 … 11 12 … 22 23

10 ms

DECT slot
synch, control, CRC data guard time
bits: 96 320 44
368 µs 49µs

Figure 2.18: Frame structure in the DECT system [11].

In each time slot one out of ten carrier frequencies is used for transmission. The
mobile telephone continuously measures the transmission quality of all the carriers
and chooses the best one for communication [11]. The TDD structure is used for a
simple pre-compensation arrangement, where antenna diversity is used as men-
tioned in the previous section. The base station has two receive antennas, which
both receive the incoming signal simultaneously. Hence some kind of space diver-
sity (antenna diversity) operation can be used on the uplink. One possible method
for the space diversity system is to use e.g. selection diversity, which is easy to
implement though it is not the optimal method [22]. The strongest signal is used for
demodulation and the diversity arrangement only consists of the extra antenna, a
sensing device and a switch. On the downlink pre-compensation is used in order to
control the output power of the transmitter. Space diversity on the down link is
achieved by using the same antenna as was used for reception. If the transfer func-
tion of the channel remains constant during the uplink and downlink block and if we
assume equal disturbances at both ends, the best antenna on the uplink will also be
the best on the downlink. Thereby is space diversity achieved both for the mobile
terminal and for the base station, even though all the diversity equipment is located
at the base station.
The performance of TDD systems are often sensitive to timing errors and especially
pre-compensated systems are sensitive to fast changes of the used radio channel.
The DECT system is therefore mainly used for pedestrian speed [22] and small
propagation delays.
2 Background 23

Appendix A

Some specifications of OFDM

systems

Below follows a list of parameters used in some OFDM systems. DAB refers to the
Digital Audio Broadcasting system in Europe [3]. DVB means the standard for the
European Digital Video Broadcasting system [2]. BRAN, Broadband Radio Access
Networks, refers to the preliminary work by an ETSI group for a wireless local area
network [17] (one of the proposals). Finally, demonstrator refers to the test-bench
built at the department of applied electronics as a co-project between some Swedish
universities. The parameters for the demonstrator are very preliminary since the
specification is not yet finalized.

Table A.1: Parameters for some OFDM systems

System DAB DVB (TV) BRAN Demonstrator

(audio) (wireless (mobile radio)
LAN)

Frequency band (41MHz- 470 MHz- 5 GHz 1.8 GHz or

223 MHz)a / 854 MHz 5 GHz
(41MHz-
1.5 GHz)b /
>3GHzc

FFT size 2048/512/ 8196d/2048e 512 1024

256

Useful sub-channels 1536/384/ 6817/1705 345 832

192

Sub-channel spacing 1/4/8 1.11/4.46 39 ≈20

(kHz)

Symbol time Ts, without 1000/250/ 896 25.6 ≈50

CP (µs) 125
24 2 Background

Table A.1: Parameters for some OFDM systems

System DAB DVB (TV) BRAN Demonstrator

(audio) (wireless (mobile radio)
LAN)

Length of cyclic prefix 24.6 25, 12.5, 3.12 12.5

(compared to Ts, %) 6.25, 3.12

Modulation π/4 D- 4-, 16-, 64- 8-DPSK 16-QAM, 8-

QPSK QAM or 4-DPSK

Frame length (symbols) 76/76/153 68 128 4-8

Coding Punctured RS-code RS- Turbo-code,

conv. code, (204,188,8) code(86,70) R=1/2
R=1/4-8/9 + +Block
Punctured code R=2/3f
conv. code
R=1/2-7/8

Useful bitrate (bits/s) 32.103- 4.98.106- 15.93.106 ≈20.106

384.103 31.67.106
a. transmission mode I
b. transmission mode II
c. transmission mode III
d. 8k mode
e. 2k mode
f. constructed from a convolutional code
Chapter 3

Paper summary

The three papers in this thesis deal with radio channel problems in conjunction with
OFDM. The papers are directed towards three different aspects:
• How to design a pilot pattern for channel estimation?
• Which sub-channel bandwidth to use?
• What can be gained if channel compensation is moved to the basestation?
The first paper is an investigation of different pilot pattern design strategies. Differ-
ent pilot patterns are compared in terms of resulting bit error rate and a design
method is given which minimizes the resulting bit error rate. It is shown that the
pilot pattern used has a great influence on the system performance. By just a rear-
rangement of the pilot symbols, the bit error rate can be reduced by a factor five or
the number of required pilots can be decreased by an order of magnitude.
The second paper addresses the problem of choosing a suitable sub-channel band-
width for a mobile OFDM system. It is attractive in terms of capacity to have nar-
rowband sub-channels, but then the symbol time gets long and there is a risk that the
channel influence changes during the symbol duration. In that case the orthogonality
between the sub-channels is lost and the sub-channels interfere with each other. In
the paper this interchannel interference is modeled as additional (to the channel
noise) additive white Gaussian noise and an optimization of the sub-channel band-
width is performed regarding the interchannel interference and the energy loss
caused by the cyclic prefix. It is concluded that the optimization can be performed
without consideration to the influence of channel estimation. A closed form expres-
sion for the bit error rate in presence of the interchannel interference is given and it
is argued that the bandwidth optimization has to be performed on a worst case
assumption of the Doppler frequency and noise levels.
Finally, the third paper is a preparatory investigation of the possibilities of building
a mobile pre-compensated communication system, where most of the channel prob-
lems are taken care of at the base station. A system is investigated where the base
station performs channel estimation and compensates the signal to be sent for the
expected impact of the channel. In such a system a simple receiver structure can be
used, with a performance similar to that of a coherent receiver. Three compensation

25
26 3 Paper summary

methods are analyzed where amplitude and/or phase are adjusted in terms of result-
ing bit error rate and increase in peak-to-average ratio. It is concluded that phase-
only compensation is preferred at low signal to noise ratios but for higher signal to
noise ratios a bit error reduction by a factor 10 can be achieved by introducing
amplitude compensation as well.
Chapter 4

Discussion

Some problems related to the radio channel are presented and discussed in this the-
sis. We have seen that the systems often have to be optimized for the actual parame-
ters in order to get efficient systems. The number of required pilot symbols can be
reduced a lot by optimizing the pilot pattern to the environment in which it will be
used. The bit error rate may be decreased and the efficiency of the system may be
enhanced if the sub-channel bandwidth is chosen according to the expected propa-
gation conditions. Less complex receivers can be used if the signal to be sent is
adjusted to the current channel impact. The above mentioned optimization problems
are just a selection of optimizations that can be performed. They are not limited to
adoption to the mobile radio channel, but the time varying radio channel may be
interpreted as other time changing processes as well. This includes for example tem-
perature variations in a wire, frequency drift in an oscillator etc.
Design of communication systems is often a question of adoption to specific
requirements. In the future this will be an even more important task since it is hard
to design systems or hardware that can cover all different needs. One has to identify
the purpose of the system and design it accordingly. In order to satisfy all the
requirements of bitrate, capacity and quality we will probably see surface covering
systems, high capacity small area systems, semi-mobile systems for high bit rates,
all designed for each specific task. A challenging problem is to get all these special-
ized systems to collaborate such that the subscribers will get an ”ever working”
communication system with high bit rates in the areas where needed, in so-called
hot spots.

4.1 New mobile communication systems

Today it looks like an ever lasting race for new and better mobile communication
systems. New services are introduced and soon we will probably have high capacity
broadband systems, not mainly designed for speech. To my opinion the focus in the
last year has been on the third generation systems, what to do after GSM. The san-
dardization work within ETSI is running and the companies and universities within
the EU-project FRAMES have recently decided which multiaccess technique they

27
28 4 Discussion

propose for a future third generation system. They resolved to follow the wideband
CDMA path which was marked out by the Japanese telecommunication company
NTT by specifying and ordering such a (test) system. However, two of the proposals
for the ETSI standard were based on OFDM, but the technique has until now found
its niche in broadcast applications. There are discussions about using OFDM for
wireless local area networks, where a simple channel equalization can be used even
though the bit rate is in the order of 20Mbit/s.
There are some common goals for the new systems independent of the technology
chosen. Often the new systems has to be flexible and support high data rates, the
systems have to be both spectrally efficient and robust. The latter is maybe one of
the most important qualities from a customers point of view. A non-working system
is worth nothing even though the spectrum efficiency is splendid and it normally
supports bit rates of 2 Mbits/s. It is important to search for robust solutions that can
work in bad circumstances, e.g. at low signal to noise ratios. Therefore I think that
the reception should rely on pilot symbols. To my opinion it is a good investment to
use a few percent of the symbols as pilot symbols in order to get the 3 dB gain in
signal to noise ratio.

4.2 Future work

From a capacity point of view it is interesting to allow higher interference levels in

the radio channel. New coding schemes, e.g. turbo codes, have been invented that
result in acceptable bit error rates for very low signal to noise ratios. However it is
doubtful whether the synchronization circuits manage to fulfil their task in such a
situation. It would be interesting to combine pilot symbol and data symbol based
synchronization to see how much the performance could be enhanced. The same
thing could be made for channel estimation in order to e.g. decrease the pilot ratio,
but the question here is whether the gain is big enough to motivate the increased
complexity.
When it comes to the determination of the sub-channel bandwidth one has to keep
in mind that in paper II it is optimized without taking oscillator impairments into
account. Frequency errors call for broader sub-channels and it would be nice to
include this effect. However, there are already a lot of parameters to keep track of
and this problem does not get easier with more parameters.
The evaluation of the pre-compensated systems is just a preparatory study. Much
work remains in order to evaluate the technique. In reality the channel estimates are
non-perfect and the resulting performance degradation has to be analyzed. The reli-
ability of channel predictions and the corresponding pre-compensations get worse
with increasing blocklength, and therefore suitable blocklengths have to be evalu-
ated for mobile pre-compensated TDD systems. The latter problem relates in some
way to the problem of determining the number of pilot symbols that have to be used
in the channel estimator in order to follow time and frequency variations of the
channel in a satisfactory manner.
Bibliography

[1] European Telecommunications Standard Institute, GSM 05.05, version 4.6.0,

July 1993.
[2] ”Digital broadcasting systems for television, sound and data services”. Euro-
pean Telecommunications Standards Institute, prETS 300 744, Valbonne,
France, 1996.
[3] ”Radio broadcasting systems; Digital Audio Broadcasting (DAB) to mobile,
portable and fixed receivers”, European Telecommunications Standards Insti-
tute, ETS 300 401, Valbonne, France, 1995.
[4] M. Alard and R. Lasalle, ”Principles of modulation and coding for digital
broadcasting for mobile receivers”, EBU Technical Review, no. 224, pp. 168-
190, August 1987.
[5] J. A. C. Bingham, ”Multicarrier modulation for data transmission: An idea
whose time has come”, IEEE communications magazine, vol 28, no. 5, pp. 5-
14, May 1990.
[6] J. K. Cavers, ”An analysis of pilot symbol assisted modulation for Rayleigh
fading channels”, IEEE Trans. Vehic. Tech. vol 40, no. 4, pp. 686-693,
November 1991.
[7] R. W. Chang, ”Synthesis of band-limited orthogonal signals for multichannel
data transmission”, Bell system technical journal, nr 45, pp. 1775-1796,
December 1966.
[8] R. H. Clarke, ”A statistical theory of mobile-radio reception”, Bell Systems
Technical Journal, vol. 47, pp. 957-1000, 1968.
[9] O. Edfors, ”Low-complexity algorithms in digital receivers”, Ph.D. thesis,
Luleå University of Technology, Luleå, Sweden, 1996.
[10] P. Frenger, ”Multicarrier modulation and coding for multichannel systems”,
licentiate thesis, Chalmers University of Technology, Göteborg, Sweden,
February 1997.
[11] J. D. Gibson, editor, The mobile communicaitons handbook, CRC-press,
Boca Raton, Fl, USA, 1996.
[12] D. Harvatin, ”Orthogonal frequency division multiplexing (OFDM) perform-
ance in a mobile radio channel”, Technical report EAS_ECE_1996_16, Uni-
versity of Colorado, Boulder, USA, December 1996.

29
30 Bibliography

[13] P. S. Henry, B.S. Glance, ”A new approach to high-capacity digital mobile

radio”, The Bell system technical journal, Vol. 60, no. 8, pp. 1891-1904,
1981.
[14] P. Hoeher, ”TCM on frequency-selective land-mobile fading channels”, in E.
Bilglieri and M. Luise, editors, Coded modulation and bandwidth-efficient
transmission, pp. 317-328, Elsevier Science Publications, Amsterdam, The
Netherlands, 1992.
[15] P. Hoeher, S. Kaiser and P. Robertson, ”Pilot-symbol-aided channel estima-
tion in time and frequency”, in K. Fazel and G. P. Fettweis, editors, Multi-
Carrier Spread-Spectrum, chapter IV, pp. 169-178, Kluwer Academic Pub-
lishers, Dordrecht, The Netherlands, 1997.
[16] W. Jakes, Microwave Mobile Communications, Wiley-Interscience, USA
1974
[17] P. Madec, D. Lacroix, D. Castelain, J. Veillard and B. Le Floch, ”A COFDM
air interface for wireless ATM networks (HIPERLAN2)”, WG3 temporary
document 33, ETSI EP BRAN, European Telecommunications Standards
Institute, Valbonne, France, 1997.
[18] A. Masoomzadeh-Fard and S. Pasupathy, ”Combined equalization and differ-
ential detection using precoding”, IEEE transactions on communications, vol.
45, no. 3, pp. 274-278, March 1997.
[19] H. Matsuki and H. Takanashi, ”Theoretical analysis of BER improvement by
adaptive transmitter power control for Rayleigh fading compensation”, Elec-
tronics Letters, 29, (17), pp. 1520-1521, 1993.
[20] R. Nilsson, O. Edfors, M. Sandell, and P. O. Börjesson, ”An analysis of two-
dimensional pilot-symbol assisted modulation for OFDM”. Proceedings of
IEEE International Conference on Personal Wireless Communications
(ICPWC’97), Mumbai (Bombay), India, pp. 71-74, December 17-19, 1997.
[21] A. Peled and A. Ruiz, ”Frequency domain data transmission using reduced
computational complexity algorithms”, Proc. IEEE Int. Conference Acous-
tics, speech, signal processing, Denver, USA, pp. 964-967, 1980
[22] J. G. Proakis, Digital communications, third edition, McGraw-Hill, New
York, USA, 1995.
[23] T. Rappaport, Wireless communications, principles and practice, Prentice
Hall, Upper Saddle River, NJ, USA, 1996.
[24] S. M. Redl, M. K. Weber and M. O. Oliphant, An introduction to GSM,
Artech House Publishers, Boston, USA, 1995.
[25] H. Rohling and T. May, ”Comparison of PSK and DPSK modulation in a
coded OFDM system, Proceedings of IEEE Vehicular Technology Confer-
ence, Phoenix, Az, USA, pp. 870-874, May 4-7, 1997.
[26] M. Russell and G. L. Stüber, ”Terrestrial digital video broadcasting for
mobile reception using OFDM”, Wireless Personal Communications, vol 2,
nr 1-2, Kluwer Academic Publishers, Dordrecht, The Netherlands, pp. 45-66,
1995
Bibliography 31

[27] M. Sandell, ”Design and analysis of estimators for multicarrier modulation

and ultrasonic imaging”, Ph.D. thesis, Luleå University of Technology, Luleå,
Sweden, 1996.
[28] Fredrik Tufvesson and Torleiv Maseng, ”Pilot assisted channel estimation for
OFDM in mobile cellular systems”, Proceedings of IEEE Vehicular Technol-
ogy Conference, Phoenix, USA, pp. 1639-1643, May 4-7, 1997.
[29] Fredrik Tufvesson and Torleiv Maseng, ”Optimization of sub-channel band-
width for mobile OFDM systems”, in D. Everitt and M. Rumsewicz, editors,
Multiaccess, mobility and teletraffic – advances in wireless networks, pp.
103-114, Kluwer Academic Publishers, Dordrech, The Netherlands, 1998.
[30] Fredrik Tufvesson, Michael Faulkner and Torleiv Maseng, ”Theoretical anal-
ysis of pre-compensation for Rayleigh fading channels in QPSK TDD sys-
tems”, Series of technical reports, no. 3, ISSN 1402-8840, Department of
Applied Electronics, Lund University, Lund, Sweden, January 1998.
[31] S. B. Weinstein and P. M. Ebert, ”Data Transmission by Frequency-Division
Multiplexing Using the Discrete Fourier Transform”, IEEE Transactions on
Communications, vol. COM-19, pp. 628-634, 1971.
[32] S. K. Wilson, ”Digital audio broadcasting in a fading and dispersive channel”,
Ph.D. thesis, Stanford University, USA, 1994.
[33] W. Zhuang and V. Huang, ”Phase pre-coding for frequency selective Rayleigh
and Rician slowly fading channels”, IEEE transactions on vehicular technol-
ogy, vol. 46, no. 1, pp. 129-142, Februari 1997.
[34] S.O. Öhrvik, Radio school, part DM1, Ericsson Radio Systems, Stockholm,
Sweden, 1993.
32 Bibliography
Paper I
Pilot assisted channel estimation
for OFDM in mobile cellular
systems

?
Paper I
 PILOT ASSISTED CHANNEL ESTIMATION FOR OFDM IN MOBILE
CELLULAR SYSTEMS
Fredrik Tufvesson and Torleiv Maseng
Department of Applied Electronics, Lund University, Box 118, S-221 00 Lund, Sweden
E-mail: Fredrik.Tufvesson@tde.lth.se
Abstract — The use of pilot symbols for channel estimation
introduces overhead and it is thus desirable to keep the number
of pilot symbols to a minimum. The number of needed pilot
symbols for a desired bit error rate and Doppler frequency is
highly dependent on the pilot pattern used in orthogonal fre-
quency division multiplexed, OFDM, systems. Five different
pilot patterns are analysed by means of resulting bit error rate,
which is derived from channel statistics. Rearrangement of the
pilot pattern enables a reduction in the number of needed pilot
symbols up to a factor 10, still retaining the same performance.
The analysis is general and can be used for performance analy-
sis and design of pilot patterns for any OFDM system.
I. INTRODUCTION
The mobile channel introduces multipath distortion of the
signalling waveforms. Both the amplitude and phase are cor-
rupted and the channel characteristics changes because of
movements of the mobile. In order to perform coherent detec-
tion, reliable channel estimates are required. These can be
obtained by occasionally transmitting known data or so called
“pilot symbols”. The receiver interpolates the channel informa-
tion derived from the pilots to obtain the channel estimate for
the data signal, see Fig. 1.
data
transmitter decision
channel
pilot estimator
pattern
receiver
Fig. 1 The pilot pattern is known both by the receiver and
the transmitter. The purpose is to minimize the number of
transmitted pilot symbols without increasing the bit error
rate.
OFDM, orthogonal frequency division multiplexing, is used
and proposed for several broadcast systems and there is a
chosen small enough to enable reliable channel estimates but
large enough not to increase the overhead too much. This paper
includes among all an algorithm of how to design a suitable
pilot estimation pattern.
In a multicarrier system there exist a unique opportunity to
determine various parts of the channel impulse response, as
opposed to a single-carrier system. It is no use to make efforts
to determine already known parts. Until now it seems like no
one has looked into the impact of the placement of the pilot
symbols for multicarrier systems. Cavers [5] made an exhaus-
tive theoretical analysis for single-carrier systems. He pointed
out that it was appropriate that 14% of the sent symbols were
pilot symbols to be able to handle large Doppler values
(fdTs=0.05). Some pilot estimation patterns for OFDM has
been presented in the literature, see e.g. [6], [7]. Comparisons
between these and the one proposed here are shown later.
II. ESTIMATION STRATEGIES
Five different pilot patterns are analyzed, see Fig. 2.
1. Measure all channels at the same time, compare to a broad-
band single-carrier system.
2. Measure the channels in increasing order, one at a time.
3. Measure the channels one at a time in a smart, but predeter-
mined, way. The measurement order is derived upon chan-
nel statistics and is optimal in the sense that the total bit
error rate is kept at a minimum each symbol time.
4. A pilot pattern presented by T. Mueller et al. [6], where the
pilots are located with equidistant spacings in time and fre-
quency.
5. A pilot pattern by P. Hoeher [7], used in [4]. The pilot sym-
bol locations are shifted one step in frequency each pilot
interval.
64
1) 2)
......
...

growing interest in using the technique for the next generation Pilot
symbol
Channel

land mobile communication system. In OFDM systems the

information signal can be seen as divided and transmitted by Data
symbol
several narrowband sub-carriers. Typically, for practical 2
...
1 ...
OFDM systems, the frequency spacing is less than the coher- 1 2 Time 65
ence bandwidth and the symbol time is less than the coherence
time. This means that a receiver and pilot estimation pattern 3) 4) 5)
...
...
...

that take advantage of the relatively large coherence bandwidth

and coherence time can manage with less pilot symbols,
thereby minimizing the overhead introduced by the pilot sym-
bols. The problem is to decide where and how often to insert ... ... ...
pilot symbols. The spacing between the pilot symbols shall be Fig. 2 Examples of pilot patterns analyzed.

1
2 Paper I

For comparison the same amount of pilots is used, one of 64
sent symbols. This means that only 1.6% extra overhead is
introduced by the pilots, but this is not sufficient for large
Doppler frequencies in some of the cases.
III. SYSTEM
At different signal to noise ratios the resulting bit error rate
from each pilot pattern is evaluated using the algorithms given
in Section VIII. A matched filter receiver and coherent BPSK
or QPSK are used. Additive white Gaussian noise is assumed
for every sub-channel. The channel has delays and Rayleigh
distributed amplitudes according to the COST 207: “Typical
Urban profile”. The complex parts of the transfer function are
assumed to change according to a first order auto-regressive
process as described in Section VI. The reason for using this
model is to get a linear system which rather easy can be hand-
led algebraically. A Kalman filter is used to estimate the fre-
For a given Doppler frequency the pilot pattern used sets
the limit for the lowest pilot density to be used, alternatively
for a given pilot density it limits the maximum Doppler fre-
quency allowed. The calculations are made at 1800 MHz using
10 kHz channel separation between 64 OFDM channels carry-
ing in total 640 ksymbols/s. No intersymbol interference, per-
fect synchronization and known Doppler frequency, fd, is
assumed. 1.6% of the sent symbols are pilot symbols and the
average bit error rate of the 64 channels is presented.
The bit error rate is degraded both by imperfect channel
estimates and noise disturbances. The pilot pattern used deter-
mines the conversion between noise limited and estimate lim-
ited region, see Fig. 5.
BER for QPSK when different pilot patterns are used, fdTs=0.001
0
10
Pattern 1
Pattern 2
Pattern 3
Pattern 4
Pattern 5
-1 Coherent
10

quency response of the channel. From the filter, a time depen-

dent covariance matrix is given as described in Section VII. BER
-2
10
This is used to calculate the expected bit error rate for each
channel. See Fig. 3 for a description of the system.
-3
10
I ”Typical urban” X
PSK channel, H +

Used for design

noise: e -4
10
0 5 10 15 20 25 30
pilot Eb/No

P, K, C channel Fig. 5 Bit error rate at different signal to noise ratios. Note
Estimated BER estimator the difference between the error floors.
Ĥ It is interesting to study the resulting pattern when estimat-
ing the channel that gives the lowest bit error rate (strategy 3).
Real BER Î decision Ĥ∗
matched filter
A steady state pilot pattern is often achieved where only few of
Fig. 3 Overview of the system used to analyze the estima- the sub-channels are measured, see Fig. 6. Channel estimates
tion patterns. of the other sub-channels are achieved by filtering.
Pilot pattern, fdTs=0.002 Pilot pattern, fdTs=0.02

IV. RESULTS 60 60

The resulting bit error rate curves of the pilot patterns are 50 50

presented for different Doppler frequencies in Fig. 4. 40 40

Channel

Channel

BER for QPSK when different pilot patterns are used, Eb/No=10
30 30

Pattern 1
Pattern 2 20 20
Pattern 3
Pattern 4
Pattern 5
Coherent 10 10

-1 20 40 60 80 100 20 40 60 80 100
10 Time Time

Fig. 6 Resulting pilot pattern when sending a pilot at the

BER

channel which gives the lowest total bit error rate. Note that
only few channels will be used for pilot symbols.
To see the effect of mismatch between the pilot pattern
design parameters and the actual parameters, the optimal pilot
patterns (pattern 3) for three Doppler values were used when
-2
moving at another speed. The actual Doppler frequency was as
10
10
-4 -3
10 10
-2 -1
10 before assumed to be known by the estimator, just to see the
fd*Ts
effect of the pilot pattern without influences of the estimation
Fig. 4 Resulting bit error rate as a function of Doppler value
when the different pilot patterns are used. The same number algorithm. The designed pilot patterns for the ”typical urban”
of pilot symbols are used for all curves. environment were also used in ”hilly terrain” and ”rural area”
Paper I 3

specified in [2] to see the influence of the propagation environ- The bit error rates within the sub-channels differ depending
ment on the bit error rate, see Fig. 7. on where the pilots are located. When minimizing the total bit
BER when using pattern designed for another Doppler freq. or channel, Eb/No=10
error rate (pattern 3) the channels of the sides get higher error
rate, see example in Fig. 8.
-1
10

0.12

0.11

0.1

BER
0.09
Rural
Hilly
Urban 0.08
fd*Ts=0.02
fd*Ts=0.002
fd*Ts=0.0002 0.07

-2
10 0.06
-4 -3 -2 -1
10 10 10 10 80
fd*Ts
60
Fig. 7 Changes in the bit error rate due to Doppler mis- 40
70
match and power delay profile mismatch. The pilot patterns 20 40
50
60
30
designed for fdTs=0.02, 0.002 and 0.0002 were used at dif- Time 0 0
10
20

ferent Doppler values and the designed patterns for typical Channel

urban environment were used in hilly terrain and rural area.
As seen in the figure, the pilot patterns are robust to mis-
matches in the design parameters.
V. DISCUSSION
To minimize the bit error rate it is desirable to spread the
pilot symbols both in time and in frequency, as seen in Fig. 4
and Fig. 5. Normally a worst case design is made for the chan-
nel estimation system and then we suggest to tailor the pilot
pattern to each base station site. A suitable pilot pattern can be
calculated once the propagation environment and maximum
expected speed in the particular cell is known. In such a sys-
tem, the pilot pattern used in the cell is among the information
transferred to the mobile when it logs into a new cell. When
designing the pilot pattern one also has to take the estimation
algorithm into account. Here the estimation algorithm is used
only for evaluation and the complexity of the used algorithm is
not a problem. In some cases the pilot pattern has to aid the
estimation algorithm to enable a less complex one. The estima-
tion algorithm used here, the Kalman filter based on an AR-
process, has no delay and the received signal can be detected
immediately, i.e. no future pilot symbols is taken into account
when making the channel estimate. If the received signal is
stored in a memory, the pilot symbols can be used in both
”time directions” and the time spacings between pilot symbols
can be increased, retaining the same performance.
The degradation due to mismatch in design parameters is
mainly caused by the estimation algorithm and therefore the
curves for different Doppler values do not differ much. When
the pilot pattern is designed for higher Doppler values than the
actual one, an increased error rate is achieved since the pilot
symbols are not located as close to each other in frequency as
desired. In rural area the bit error rate becomes lower due to the
increased frequency correlation while the opposite happens in
hilly terrain. In the first case, an even better result is achieved
with less pilots along the frequency axes and more along the
time axes.
Fig. 8 The bit error rate becomes higher for the side chan-
nels when the total bit error rate is kept at a minimum
VI. CHANNEL MODEL
The time dependent impulse response, h ( τ, t ) , is assumed to
be a sum of reflections, see (1) where δ ( t ) denotes the dirac-
function.
N
h ( τ, t ) = ∑ an ( t ) ⋅ δ ( t – τn ) (1)
n=1
The tap coefficients, an(t), and the tap delays, τ n , are chosen
according to the COST 207 ”Typical Urban” model in the
GSM specification [2]. The transfer functions, H(f,t), are
obtained by the Fourier transform and these are the functions
we want to estimate for the different carriers. These channel
transfer functions are regarded as flat fading and constant dur-
ing a symbol time.
A first order AR-process is used to model how the different
taps may change from one time instant to another. If we look at
all the transfer functions at the same time, it is possible to set
up a state-space model of the form:
H ( k + 1 ) = φH ( k ) + v ( k )
(2)
y ( k ) = C ( k )H ( k ) + e ( k )
The matrix φ is a diagonal N*N-matrix (here treated as a sca-
lar) with elements
– k AR 2πf d T s
e (3)
that define the AR-process. Ts is the symbol time including any
cyclic prefix or guard space. The white noise v(k) has covari-
ance matrix R 1 = F ⋅ RGSM ⋅ F∗ , where RGSM corresponds to the
multipath intensity profile described in [2]. The vector y(k) is
the measured transfer functions. C(k) is an observation vector
with ones only at the positions (channels) measured at time KTs
and e(k) is measurement noise with a diagonal covariance
matrix R2.
The parameter kAR in the auto-regressive process for the
channel changes is chosen to adjust the ”memory” so that it
4 Paper I

corresponds to the ”memory” of Jakes’ model. Channels corre-

K ( k ) = φP ( k k – 1 )C ( k )∗ (6)
sponding to the U-shaped spectrum given in [8] were simulated
–1
and then estimators based on an AR-model with different kAR ⋅ [ C ( k )P ( k k – 1 )C ( k )∗ + R 2 ]
were used, see Fig. 9
P ( k + 1 k ) = φP ( k k – 1 )φ∗ + R 1 (7)
– K ( k ) [ C ( k )P ( k k – 1 )C ( k )∗ + R 2 ]K ( k )∗
Simulated BER for QPSK, estimates based on AR-models with different fdTs
0.15

data symbols 1-9

mean
0.14
P ( k k ) = P ( k k – 1 ) – P ( k k – 1 )C ( k )∗ (8)
–1
0.13 ⋅ [ C ( k )P ( k k – 1 )C ( k )∗ + R 2 ] C ( k )P ( k k – 1 )
The reconstruction error H̃ ( k k ) = H ( k ) – Ĥ ( k k ) is given by:
BER

0.12

0.11
H̃ ( k k ) = [ φ – K ( k )C ( k ) ]H̃ ( k – 1 k – 1 ) + v ( k – 1 ) – (9)
–1
P ( k k – 1 )C ( k )∗ [ C ( k )P ( k k – 1 )C ( k )∗ + R 2 ]
0.1

0.09
10
-1
10
0 1
10
fdTs used in estimator (fdTs of estimator)/(fdTs of simulated channel)
Fig. 9 Bit error rate for a simulated channel by Jakes when
estimates are based on an AR-process. The minimum value
is reached for kAR=0.15.
Simulations were performed with one sub-channel (Eb/No= 10,
fdTs=0.002) with every tenth symbol as a pilot symbol. In the
figure the bit error rates of the nine data symbols are shown.
Fig. 9 shows that the best fit, in this case, is reached for
kAR≈0.15.
The adjustment of the AR-process can also be seen as an
adjustment of the bandwidth. If we set the 90% power band-
width of the AR process equal to the bandwidth of the Doppler
spread, see Fig. 10, a value of kAR≈0.158 can be calculated.
This is the value used for all bit error rate calculations through-
out the paper.
Doppler power spectrum of AR-process and channel by Jakes
S(f)
⋅ { e ( k ) + C ( k )v ( k – 1 ) }
The Kalman filter is optimal in the sense that the variance of
the reconstruction error is minimized. The matrix P(k|k) is the
variance matrix and this is used together with K to make an
estimate of the bit error rate, which in turn is used to decide the
order in which the channels are going to be measured. For pat-
tern 3, the channel is chosen that minimizes the total bit error
rate of the channels after the measurement. The matrices P(k|k)
and K are independent of the measured values and therefore it
is possible to precompute the order in which the channels are
going to be measured.
VIII. BIT ERROR RATE CALCULATIONS
The bit error rate is calculated for BPSK and QPSK. A
matched filter receiver is assumed. The sampled output of this
filter is given by
2π
j ------ ( n – 1 )
M
X m = 2εH m e + em (10)
where ε is the signal energy, Hm is the current transfer function

at channel m, n is the signal alternative among M sent and em is

white gaussian noise. The bit error rate at channel m is calcu-
lated as [1]
1
-2 -1.5 -1 -0.5 0
frequency (f-fc)/fd
0.5 1 1.5 2
P bmBPSK = --- ( 1 – µ m )
Fig. 10 Power density spectrum of the channel by Jakes and 2
a first order AR-process. The latter is adjusted to have the  µ m  (11)
P bmQPSK = ---  1 – -
90% power bandwidth equal to the Doppler spread. 1
-- - - - - -
2 2
 2 – µ m
VII. CHANNEL ESTIMATOR
where
For the analysis and pilot pattern design a Kalman filter is
used to estimate the transfer functions. The Kalman filter is E [ X m Hˆm∗ ]
causal and uses measurements up to time k to estimate the µ m = ------------------------------------------------- (12)
2 2
transfer functions, H(k). The Kalman filter is given [3] by (4)- E [ X m ]E [ Hˆm ]
(8), where X∗ denotes conjugate transpose of X:
Hˆm are the outputs of the Kalman filter. These are not known in
Ĥ ( k k ) = Ĥ ( k k – 1 ) + P ( k k – 1 )C ( k )∗ (4) advance and therefore (13) - (15) are used. If matrix notation is
–1 used and signal alternative 1 is used for the pilot symbol (does
⋅ { C ( k )P ( k k – 1 )C ( k )∗ + R 2 } ( y ( k ) – C ( k )Ĥ ( k k – 1 ) ) affect the analysis here, but in practise different pilot symbol
Ĥ ( k + 1 k ) = φĤ ( k k – 1 ) + (5) alternatives should be used), the expectations for all the chan-
nels can be calculated as:
K ( k ) [ y ( k ) – C ( k )Ĥ ( k k – 1 ) ] = φĤ ( k k )
Paper I 5

E [ X ( k )Ĥ ( k )∗ ]
surement is useful. In a similar way, the bit error rate when
(13)
estimating a constant channel with use of (infinitely) many
= E [ { 2εH ( k ) + N ( k ) } { H ( k ) – H̃ ( k ) }∗ ]
pilot symbols corresponds to that of coherent detection. Nor-
= 2εE [ H ( k )H ( k )∗ ] – 2εE [ H ( k )H̃ ( k )∗ ] mally in a practical system the effect of the estimation is some-
E [ X ( k ) ] = E [ { 2εH ( k ) + N ( k ) } { 2εH ( k ) + N ( k ) }∗ ] (14)
2 where between these two cases.
= 4ε E [ H ( k )H ( k )∗ ] + E [ N ( k )N ( k )∗ ]
2
IX. CONCLUSIONS
E [ Ĥ ( k ) ] = E [ { H ( k ) – H̃ ( k ) } { H ( k ) – H̃ ( k ) }∗ ]
2
(15) The bit error rate for pilot assisted QPSK modulation is cal-
= E [ H ( k )H ( k )∗ ] – 2E [ H ( k )H̃ ( k )∗ ] + E [ H̃ ( k )H̃ ( k k )∗ ]
culated when using different pilot patterns. The ability to esti-
mate the channel reliably when it changes due to e.g. vehicle
The expected values E [ X ( k ) Ĥ ( k )∗ ] , E [ X ( k ) 2 ] and E [ Ĥ ( k ) ] are
2
movements is highly dependent on the pilot pattern used. By
independent of the measured values but dependent upon which rearranging the pilot pattern it is, in some cases, possible to
channels that are measured. Therefore they are time dependent. handle 10 times higher Doppler frequency alternatively possi-
The expectations can be calculated as: ble to reduce the number of needed pilot symbols the same
amount, still retaining the same bit error rate. Alternatively the
E [ H ( k )H ( k )∗ ] (16)
new pilot pattern could be used to reduce the bit error rate up to
= E [ { φH ( k – 1 ) + v ( k – 1 ) } { φH ( k – 1 ) + v ( k – 1 ) }∗ ] a factor 5, even more in a low noise environment. The pilot
= φ E [ H ( k – 1 )H ( k – 1 )∗ ] + E [ v ( k – 1 )v ( k – 1 )∗ ]
2
symbols in the proposed pilot pattern are spread out both in
= R1 ⁄ ( 1 – φ )
2 time and frequency. For a given propagation environment, e.g.
a base station site, it is possible to pre-calculate a suitable pilot
E [ N ( k )N ( k )∗ ] = R 2 (17) pattern.

E [ H ( k )H̃ ( k k )∗ ] = E [ H ( k )H ( k )∗ – H ( k )Ĥ ( k k )∗ ] (18) REFERENCES

= E [ φH ( k – 1 )H̃ ( k – 1 )∗ φ∗ + R 1 – R 1 C ( k )∗ a∗
– φH ( k – 1 )H̃ ( k – 1 )∗ φ∗ C ( k )∗ a∗ ]
[1] J. G .Proakis, Digital Communications, third edition,
McGraw-Hill, USA, 1995.
= ( φE [ H ( k – 1 )H̃ ( k – 1 )∗ ]φ∗ + R 1 ) ( I – C ( k )∗ a∗ )
[2] European Telecommunications Standard Institute, GSM
where 05.05, version 4.6.0, July 1993.
E [ H̃ ( k k )H̃ ( k k )∗ ] = P ( k k ) (19) [3] J. M. Mendel, Lessons in Estimation Theory for Signal Pro-
cessing, Communications and Control, Prentice-Hall, USA,
–1
a = P ( k k – 1 )C ( k )∗ { C ( k )P ( k k – 1 )C ( k )∗ + R 2 } (20) 1995.
The bit error rate calculation is compared and verified by [4] M. Sandell, Design and Analysis of Estimators for Multi-
simulations. Proakis [1] gives expressions for the bit error rate carrier Modulation and Ultrasonic Imaging, Ph.D. thesis, Luleå
when estimating a constant Rayleigh channel using different University of Technology, Sweden, 1996.
numbers of pilot symbols, see Fig. 11. [5] J. K. Cavers, An Analysis of Pilot Symbol Assisted Modu-
BER for BPSK, constant Rayleigh channel, Eb/No=0, 10, 20 and 30
lation for Rayleigh Fading Channels, IEEE Trans. Vehic. Tech.
0
10
Proakis
vol 40, no 4, pp 686-693, nov 1991.
coherent
own calculations
own simulations
[6] T. Mueller, K. Brueninghaus and H. Rohling, Performance
-1
of Coherent OFDM-CDMA for Broadband Mobile Communi-
10
cations, Wireless Personal Communications 2, pp 295-305,
Kluwer, Netherlands, 1996.
[7] P. Hoeher. TCM on Frequency-Selective Land-Mobile Fad-
BER

-2
10
ing Channels, Proc. of 5th Tirrenia Int. Workshop in Dig. Com-
munications, Tirrenia, Italy, Sept. 1991.
-3
10 [8] W. Jakes, Microwave Mobile Communications, Wiley-
Interscience, USA 1974

-4
10
0 2 4 6 8 10 12 14 16 18 20
Number of pilot symbols used for the estimate

Fig. 11 Bit error rate for 2PSK when estimating a constant

Rayleigh channel with several pilot symbols at different
signal to noise ratios.
The bit error rate when using only one pilot symbol corre-
sponds to estimating a rapidly changing channel or the first
estimate of an unknown channel. Then, only the latest mea-
6 Paper I
Paper II
Optimization of sub-channel
bandwidth for mobile OFDM
systems
Paper II
 OPTIMIZATION OF SUB-CHANNEL
BANDWIDTH FOR MOBILE OFDM
SYSTEMS

Fredrik Tufvesson and Torleiv Maseng

Department of Applied Electronics, Lund University,
Box 118, SE-221 00 Lund, Sweden.
E-mail: Fredrik.Tufvesson@tde.lth.se

Abstract: The bit error rate in orthogonal frequency division multiplex (OFDM) systems
is affected by the number of sub-channels used, changes in the channel characteristics and,
to some extent, the excess delay of the channel. This paper presents an analytical expres-
sion for the bit error rate on Rayleigh fading channels when interchannel interference (ICI)
caused by channel changes during a symbol and energy loss due to the cyclic prefix are
regarded. This expression is used to optimize the number of sub-channels, and thereby the
sub-channel bandwidth in the system. It is argued that the system can be optimized
neglecting the effect of imperfect channel estimation and on a worst case assumption for
the Doppler frequency and signal to noise ratio. The analysis is general and can be used for
performance analysis and optimization of any mobile OFDM system.

INTRODUCTION

OFDM, orthogonal frequency division multiplexing, is used and proposed for several
broadcast systems [1] [2] and there is a growing interest in using the technique for the
next generation land mobile communication system. In OFDM systems the informa-
tion signal can be seen as divided among and transmitted by several narrowband sub-
carriers. The bandwidths of the sub-carriers depend on the bit rate sent on each of
them and consequently, for a given total bitrate the sub-carrier bandwidth is depen-
dent on the number of sub-channels used. The number of sub-channels used is set in
the system design and the problem is to find a good trade off between bandwidth,
limitations by the hardware and the physical channel. As the sub-carrier bandwidth is
reduced, the symbol time on each carrier gets longer, the channel changes during a
symbol gets larger and channel compensation gets more difficult. When the changes
in the channel characteristics during a symbol time become evident, the orthogonality
between the subchannels is lost and interchannel interference (ICI) arises. On the

1
2 Paper II

other hand as the carrier bandwidth and the sub-channel symbol rate increases, inter-
symbol interference (ISI) becomes a problem. This problem can be avoided by intro-
ducing a cyclic prefix [3], but this results in an energy loss. In between these two
extremes there exists an optimal bandwidth where the bit error rate is minimized, this
bandwidth is found in this paper.
There is some previous work regarding the choice of number of sub-channels in
OFDM systems, see e.g. [4] [5], but often the optimization is performed empirically
or by simulation. In [4] a noncoherent OFDM DPSK system is analysed with respect
to random FM noise and a frequency selective channel. It is concluded that a sub-
channel bandwidth of 1.5 kHz is suitable for a bitrate of 200 kbit/s and 100 Hz Dopp-
ler frequency, but without consideration to the interchannel interference caused by
channel changes during a symbol. The latter can cause severe degradation of the bit
error rate if the sub-channel bandwidth is chosen too small.

SYSTEM DESCRIPTION

When designing an OFDM system one often start with a requirement on the total bit
rate. Assume that a total bitrate of Rtot bits per second is required. If these are equally
divided between the channels then the bitrate

R tot
R 1 = --------- (1)
N

is transferred on each of the N sub-channels. The symbol time for M-ary modulation
on each sub-channel becomes

log 2M N log 2M
T sub = --------------- = -------------------- . (2)
R1 R tot

In the following the influence of some of the most important design parameters are
discussed.

Cyclic prefix

To avoid intersymbol interference and to maintain orthogonality between the sub-

channels in a time dispersive channel, a cyclic prefix can be used [3]. ISI is avoided if
the length of the cyclic prefix, Tcp, is chosen large enough to exceed the maximum
excess delay of the channel. The total symbol time, Tsub, is extended when a cyclic
prefix is used but the sub-channel bandwidth, ∆f , is equal to the inverse of the sym-
bol time excluding the cyclic prefix [4], Ts, see Figure 1.

1 1 R tot
∆f = ----- = -------------------------- = ------------------------------------------- (3)
Ts T sub – T cp N log 2M – T cp R tot
Paper II 3

copy
f

N*∆f

t
∆f

Tcp Ts
Tsub

Figure 1. Eight symbols sent on four sub-channels. The striped area indicates
insertion of the cyclic prefix.
Insertion of a cyclic prefix means that the sub-channel bandwidth has to be increased
in order to keep the bit rate R1 constant. The bit rate determines the symbol time Tsub
and if a part of this is used for the cyclic prefix, then the time Ts has to be shortened
which in turn leads to increased sub-channel bandwidth.
The cyclic prefix also leads to a power loss, α cp . The receiver uses the energy
received during the time Ts and discards the energy corresponding to the duration of
the cyclic prefix. The remaining signal energy can be calculated as α cp E s where

Ts T cp R tot
α cp = ----------- = 1 – -------------------- . (4)
T sub N log 2M

Interchannel interference

Even though the channel is perfectly estimated there are some variations in the trans-
fer function during each symbol interval. These variations becomes evident when
many sub-channels are used due to the long symbol time, they are hard to track and
result in interchannel interference, ICI. When the number of sub-channels is suffi-
ciently large the resulting ICI can be modeled as additive white Gaussian noise,
which is added to the channel noise [6] with spectral density N0. The variance of the
ICI-noise, σ 2ICI = E [ n ICI n ICI∗ ] , is calculated as[6]:

N–1
 log 2M T cp 
σ ICI = 2E s  1 – ---- – ------2- ∑ ( N – n )J 0 2π f D n  --------------- – --------  (5)
2 1 2
 N N  R tot N 
n=1
4 Paper II

Bit error rate for coherent detection

If we neglect the effect of channel estimation errors, the equivalent signal to noise
ratio after the cyclic prefix is removed and the ICI-noise is added becomes

E E s α cp
 ------s- = -----------------------
-. (6)
 N 0 eq 2
N 0 + σ ICI

The BER for QPSK/4QAM in a Rayleigh fading channel with coherent detection can
then be calculated as

∞
 Es 
p ( y )Q  y  ------  dy = --- – ---
1 1
pb = ∫   N 0 eq 2 2
(7)
y=0

N E s ( 2 N – T cp R tot )
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
N–1
log 2M T cp
N ( 4 N 0 + 10E s ) – N E s ( 8 + T cp R tot ) – 16E s ∑ ( N – n ) J 0 2π f D n  --------------
2
- – --------
 R tot N
n=1

where y is the power attenuation of the channel with an exponential distribution of

mean 1, i.e. p(y)=e-y. In Figure 2 the bit error rate for coherent detection is presented
for four different Es/N0 at 50 Hz Doppler frequency and a total bit rate of 320, 640
respectively 1280 kb/s. The reference curve represents coherent detection in a Ray-
leigh channel without losses due to cyclic prefix or ICI.

3 dB
-1
10
13 dB

-2
10
BER

23 dB

-3
10
33 dB
loss incl.
no loss
-4
10
3 4 5
10 10 10
Sub-channel bandwidth/Hz

Figure 2. The BER curves for Rtot=320, 640 and 1280 kbit/s overlap each other
totally in this case. Coherent QPSK is used in a Rayleigh fading channel with
Tcp=10 µs and fD=50 Hz when the number of sub-channels and signal to noise
ratio are varied. The dotted lines represent coherent detection without losses.

A cyclic prefix of duration 10 µs is chosen to combat time dispersion, a duration long

enough to exceed the expected maximum excess delay of the channel. In this paper a
COST 207 ”Typical Urban” channel [8] is considered with an maximum excess delay
of 5 µs. In Figure 2 the bit error rate is shown insensitive to bitrate changes but the
effect of the ICI-noise and the energy loss due to the cyclic prefix is apparent. On the
right hand side few sub-channels are used, the symbol time is short and the length of
Paper II 5

the cyclic prefix needed to suppress ISI causes a big loss of symbol energy. On the
left hand side the symbol time is long, the changes in the channel characteristics dur-
ing a symbol becomes significant and ICI arises. The optimal sub-channel bandwidth
is sensitive to the Doppler frequency since a large Doppler frequency results in large
ICI-noise and the optimum is pushed toward broader channels, see Figure 3.

3 dB
-1
10
13 dB

-2
10
23 dB
BER

-3
10 33 dB

-4
10 10 Hz
50 Hz
200 Hz
-5
10
3 4 5
10 10 10
Sub-channel bandwidth/Hz

Figure 3. Bit error rate for coherent detection when the Doppler frequency is var-
ied, fD={10, 50, 200} Hz, Rtot=320 kbit/s and Es/N0= {3, 13, 23, 33} dB. The dot-
ted lines represent coherent detection without losses.

Channel estimation

The radio channel corrupts the transmitted signal and in order to make coherent
detection possible we have to know the impact of the channel. Both the amplitude
and phase are corrupted by the fading channel, whose characteristics vary because of
movements of the mobile terminal. In order to keep track of the channel characteris-
tics pilot symbol assisted modulation, PSAM, can be used. This means that known
training symbols are multiplexed into the data stream at certain sub-channels and cer-
tain times. The receiver interpolates the channel information derived from the pilot
symbols and makes channel estimates for the data symbols. Since the pilot symbols
carry no data the pilot density is to be kept at a minimum not to increase the overhead
too much. In order to make it possible for the receiver to achieve nearly maximal
channel information from each of the pilots the pilot pattern can be made ”balanced”
[7]. This means that the spacings between the pilots are approximately the same in
both frequency and time when normalized by the minimum expected coherence
bandwidth and the minimum expected coherence time respectively. Few sub-chan-
nels means that we have to insert several data symbols between the pilot symbols in
the time domain. Many sub-channels gives us the opportunity to use several of the
channels for pilots simultaneously without increasing the pilot density, see Figure 4.
6 Paper II

f f f f

t t t t
N=1 N=2 N=4 N=8
Pilot symbols Data Symbols

Figure 4. Time - frequency distribution of a pilot symbol differs depending on the

number of sub-channels, N, used. Note that the same pilot density, 1/4, is used for
the three alternatives.
Let NT and NF denote the pilot time distance respectively pilot frequency distance in
number of symbols, fDmax denote the maximum expected Doppler frequency (Hz)
and τ max denote the maximum expected excess delay (s) of the channel, then the con-
dition for ”balanced design” becomes [7]

f Dmax ⋅ T sub ⋅ N T ≈ τ max ⋅ ∆f ⋅ N F . (8)

1
The overall pilot density is β ≈ --------------- , and as a rule of thumb the pilot spacings can be
NT NF
chosen as [7]

1
f Dmax ⋅ T sub ⋅ N T = τ max ⋅ ∆f ⋅ N F = --- (9)
4

An example of a pilot pattern with NF=4 and NT=5 is given in Figure 5.

NT
f

Data

Pilot
NF

Figure 5. An example of a ”balanced” pilot pattern with pilot time distance NT=5
and pilot frequency distance NF=4.

With a fixed time-frequency block, edge problems can arise when designing the pilot
pattern. In order to get rid of this problem in the following and in order to suppress
the bit error rate at the edges, pilots are put on the outer sub-channels respectively
time instants and then sufficiently many pilot symbols are evenly distributed between
these edge pilots in frequency respectively time in order not to exceed the pilot dis-
tances given by (9).
To see the effect of the channel estimation only, the bit error rate was calculated with-
out ICI-noise and energy losses due to the cyclic prefix. For the estimation a two-
dimensional Wiener filter [7] was used, which is optimal in the sense that the variance
Paper II 7

of the estimation error is minimized. Furthermore, known Doppler frequency, perfect

synchronization, a COST 207 ”Typical Urban” channel [8] and a matched filter
receiver was assumed, see Appendix A for more details. The resulting bit error rate
under these assumptions with a bit rate of 320 kbit/s, 200 Hz Doppler frequency is
presented in Figure 6.

3 dB
-1
10
13 dB

-2
10
BER

23 dB

-3
10 33 dB

-4
10
3 4 5
10 10 10
Sub-channel bandwidth/HZ

Figure 6. Changes in the bit error rate caused by the pilot pattern, Es/N0={3, 13,
23, 33} dB. As seen, the pilot pattern has virtually no effect on the bit error rate
when ”balanced design” is used.

The reason why the different pilot patterns virtually do not affect the bit error rate is
explained by the fact that the pilot symbols are placed at approximately the same fre-
quencies and time instances independent of the sub-channel bandwidth and symbol
time used. There are some changes in the bit error rates due to the fact that the pilot
distances can not be below one, due to numerical reasons in the pilot pattern design
and due to the slight increase in the total bandwidth to estimate when few channels
are used, but these changes are rather small.

COMPLETE SYSTEM

In a real system the channel estimation is of course not perfect. The bit error rate
when both channel estimation and the losses due to the cyclic prefix and ICI-noise are
included is lower bounded by the bit error rate for coherent detection in (7) and the
deviation gets smaller as the pilot density increases. The bit error rate for the same
parameters as in last section (Rtot=320 kbit/s fD=200 Hz, ”Typical Urban” channel,
Tcp=10 µs) but with the losses included is presented in Figure 7.
8 Paper II

3 dB
-1
10 13 dB

-2
10 23 dB
BER

-3
33 dB
10
est. incl.
coherent
-4
10
3 4 5
10 10 10
Sub-channel bandwidth/Hz

Figure 7. Bit error rate for coherent detection and for the total system when chan-
nel estimation errors are included, Es/N0={3, 13, 23, 33} dB.

DISCUSSION

As indicated in Figure 7 it is possible to optimize the number of sub-channels and

thereby the sub-channel bandwidth by assuming perfect channel estimates. The
resulting optimum when channel estimation effects are included is principally unaf-
fected by the number of sub-channels used. The reason for this is that the variance of
the estimation error, see Appendix A, is approximately the same, independent of the
sub-channel values chosen. This, in turn, means that it is possible to optimize the
number of channels neglecting the effect of channel estimation.
More narrowband sub-channels can be used when terminal movements are slow since
the channel is not expected to change during the longer symbol time. This means that
the relative duration of the cyclic prefix becomes shorter and that the bandwidth
expansion is kept small. However, the sub-channel bandwidth has to be optimized on
a worst case assumption for the Doppler frequency, otherwise the increase in bit error
rate can become severe for high Doppler values.
The optimal sub-channel bandwidth is unfortunately dependent on the signal to noise
ratio. Higher noise levels call for more narrowband sub-channel since the noise
makes the energy loss due to the cyclic prefix more evident. In the same way as for
the Doppler frequency a worst case design has to be made with respect to the noise
level. It is important to keep the maximal bit error rate as low as possible, even
though this results in a larger BER increase for lower noise levels. Actually, often it is
the performance of a coded system that is interesting and since the coding gives more
distinct optima it is important not to increase the worst case BER.
The effect of frequency offsets is not included in the analysis and this effect becomes
more significant when many narrowband sub-channels are used. The carrier to inter-
ference ratio due to frequency offsets is proportional to ∆f 2 [9], thus moving the opti-
mal sub-channel bandwidth towards broader channels when frequency offsets are
present. Finally, the calculations are based on the assumption that the sub-channels
are flat fading, which may not be the case when broadband sub-channels are used in
some channels, e.g. hilly terrain.
Paper II 9

CONCLUSIONS

The impact of the number of sub-channels, and thereby the sub-channel bandwidth,
used in a mobile OFDM-system was analysed. An analytical expression for the bit
error rate in Rayleigh fading channels was presented where the impact of the cyclic
prefix and the channel changes during a symbol time are encountered. This BER
showed an optimum for the sub-channel bandwidth to be used depending on Doppler
frequency, length of cyclic prefix and signal to noise ratio. The effect of channel esti-
mation on the optimum was also examined and this showed that the optimization can
be performed on the coherent system, without taking the channel estimation into
account. The excess delay of the channel, high noise levels and efficient usage of the
spectrum call for many narrowband sub-channels while a rapidly moving terminal
call for the use of fewer more broadband sub-channels. Finally it was argued that the
sub-channel bandwidth has to be designed on a worst case assumption for the Dopp-
ler frequency and the signal to noise ratio.

Appendix A. System models

In the following section the channel model, estimation algorithm and bit error rate
calculations are presented in detail.
The time dependent channel impulse response, h ( τ, t ) , is assumed to be a sum of
reflections, see (A.1) where δ ( t ) denotes the dirac-function.

N
h ( τ, t ) = ∑ an ( t ) ⋅ δ ( t – τn ) (A.1)
n=1

The tap coefficients, an(t), and the tap delays, τ n , are chosen according to the COST
207 ”Typical Urban” model in the GSM specification [8]. The transfer functions,
H(f,t), are obtained by the Fourier transform of (A.1) and these are the functions we
want to estimate for the different carriers. These channel transfer functions are
regarded as flat fading.
For the estimation and analysis a two-dimensional Wiener filter [7] is used, which is
optimal in the sense that the variance of the reconstruction error is minimized. The
sampled signal from the matched filter receiver of a pilot symbol at channel m in time
k is given by

2π
j ------ ( p – 1 )
k k M k
Xm = Es H m e + nm (A.2)

where Es is the signal energy, p is the signal alternative sent among the M possible
ones and n is white Gaussian noise. The noise terms have a variance given by (7), but
the covariance matrix R ke = E [ n k ( n k )∗ ] is not diagonal due to the correlation between
the channels introduced by the ICI. The filter coefficients are given by

k k T –1 T
ωm = ( ( θm ) φ ) (A.3)
10 Paper II

where φ = E [ X X∗ ] is the covariance matrix for the received pilot symbols and
k k k ∗
θ m = E [ H m ( X m ) ] is a cross-covariance vector between the transfer function to be esti-
mated and the received pilot symbols. The estimates of the channel transfer functions
can then be calculated as

k k T
Ĥ m = ( ω m ) X (A.4)

where X is a vector of the sampled values from all of the pilot symbols.
Finally, the mean square error of the estimates is calculated as

k 2 k T –1 k ∗
J m = σ H – ( θm ) φ ( θm ) (A.5)

and this matrix is in turn used to calculate the QPSK bit error rate for each of the
symbols. The BER is given by [10] [11]

1 µ
P b = --- – ---------------------- (A.6)
2 2
2 2–µ

where

2
σH
µ = E s --------------------------------------------------- . (A.7)
2 2 2
( σ H + σe ) ( σ H + J )
Paper II 11

References

[1] ”Digital broadcasting systems for television, sound and data services”. European telecom-
munications standards institute, prETS 300 744, Valbonne, France, 1996
[2] ”Radio broadcasting systems; Digital Audio Broadcasting (DAB) to mobile, portable and
fixed receivers. European telecommunications standards institute, ETS 300 401, Valbonne,
France, 1995
[3] A. Peled and A. Ruiz, ”Frequency domain data transmission using reduced computational
complexity algorithms”, Proc. IEEE Int. Conference Acoustics, speech, signal processing,
Denver, USA, pp 964-967, 1980
[4] S. Hara, M. Mouri, M. Okada and N. Morinaga, ”Transmission performance analysis of
multi-carrier modulation in frequency selective fast Rayleigh fading channel”, Wireless Per-
sonal Communications, nr 2, Kluwer Academic Publishers, Dordrecht, Netherlands, pp 335-
356, 1996
[5] M. Wahlqvist, C. Östberg, J.-J. van de Beek, O. Edfors and P. O. Börjesson, ”A conceptual
study of OFDM-based multiple access schemes: Part 1 - Air interface requirements. Technical
report Tdoc 116/96 ETSI STC SMG2 meeting no 18, Helsinki, Finland, May 1996
[6] M. Russell and G. L. Stüber, ”Terrestrial digital video broadcasting for mobile reception
using OFDM”, Wireless Personal Communications, vol 2, nr 1-2, Kluwer Academic Publish-
ers, Dordrecht, The Netherlands, pp 45-66, 1995
[7] P. Hoeher, S. Kaiser and P. Robertson, ”Pilot-symbol-aided channel estimation in time and
frequency”, In K. Fazel and G. P. Fettweis, editors, Multi-Carrier Spread-Spectrum, chapter IV,
pp 169-178. Kluwer Academic Publishers, Dordrecht, The Netherlands, 1997.
[8] European Telecommunications Standard Institute, ”GSM 05.05”, version 4.6.0, July 1993.
[9] Y. Zhao and S.-G. Häggman, ”Sensitivity to Doppler shift and carrier frequency errors in
OFDM systems -- The consequences and solutions”, IEEE Vehicular Technology Conference,
Atlanta, USA, April 29 - May 2, pp. 1564-1568, 1996.
[10] S. K. Wilson, ”Digital audio broadcasting in a fading and dispersive channel”, Ph.D. the-
sis, Stanford University, USA, 1994.
[11] J. G. Proakis, ”Digital communications”, third edition, McGraw-Hill, New York,
USA, 1995.
12 Paper II
Paper III
Theoretical analysis of pre-
compensation for Rayleigh fading
channels in QPSK TDD systems
Paper III
 Theoretical Analysis of Pre-Compensation for Rayleigh
Fading Channels in QPSK TDD Systems
Fredrik Tufvesson, Michael Faulkner and Torleiv Maseng

Abstract – In time division duplex systems the uplink channel estimate can be used for com-
pensating the channel impact on the downlink symbols. Three different methods where
amplitude and/or phase are adjusted are analyzed in terms of performance limits. Closed-
form expressions for the QPSK BER are given assuming a fully known channel. Pre-com-
pensation (precoding) is an attractive alternative to differential decoding. Phase-only com-
pensation is preferred at low Eb/No, because there is no increase in transmitter dynamics,
while at high Eb/No an order of magnitude improvement in the BER can be obtained by
including amplitude pre-compensation.

I. Introduction
For digital radio transmission either differential or coherent detection can be used. Differen-
tial detection can be used with a low complexity receiver, but has a reduced performance
compared to coherent schemes. Coherent detection results in lower bit error rates, BER, but
the receiver becomes complex due to the need for channel estimation. In time division
duplex, TDD, systems the same radio channel is used both for uplink and downlink trans-
mission. This leads to the possibility of using an uplink channel estimate to pre-compensate
(also called precode) the symbols to be transmitted on the downlink. This is performed by
filtering the signal to be transmitted by the inverse channel response. Then the received sig-
nal can be demodulated directly without channel estimation nor equalization nor coherent
carrier recovery. This means that a simple receiver structure can be used (similar complexity
to differential detection), while still maintaining the performance benefits of coherent detec-
tion [1]. This also applies to systems using multilevel modulation, such as 16- and 64-QAM,
which normally require channel estimation in the mobile terminal.

In [2] an amplitude pre-compensated noncoherent FSK system was analyzed and this
showed a possible reduction of 3 dB in transmitted power at high SNR values, or a reduction
on BER by a factor five. In this paper the theoretical performances of three pre-compensated
QPSK TDD systems are compared when the channel is assumed to be known. In a practical
system this is of course not true and the results presented here are best case. But, if the chan-
nel is assumed to be reciprocal and the blocklength in the TDD structure is short compared
to the fading period, then reliable channel estimates can be made and a pre-compensated
system is possible.

II. System
The channel is described by a phase shift, θ , and a power gain, y. The received signal may
be represented by [3]

jθ ( t )
r(t ) = y ( t )e s(t ) + n(t ) (1)

where s(t) is the transmitted signal and n(t) is white Gaussian noise. In the following a flat
Rayleigh fading channel which is constant during a symbol time is considered.

Three compensation methods are evaluated:

1
2 Paper III

1. Phase compensation only, the transmitter phase is altered by – θ radians.

2. Phase and amplitude compention when y>a2, otherwise no transmission. The
transmitter output is inhibited when the path attenuation exceeds the thresh-
old a2.
3. Phase and amplitude compensation when y>a3, otherwise power limiting.
The transmitter output saturates when the path attenuation exceeds the thresh-
old a3.
The base station makes estimates of the channel amplitude and channel phase e.g. by use of
known pilot symbols on the uplink. Then the signal to be sent is compensated. If we let
u m' ( t ) denote the uncompensated equivalent lowpass waveforms [3], then the equivalent
lowpass waveforms for the compensated sent signals are

– jθ
( um ( t ) )1 = e u m' ( t ) (2)

0 y < a2

( u m ( t ) ) 2 =  η 2 – jθ y ≥ a2 (3)
 ------ e u m' ( t )
 y

 η 3 – jθ
 ------ e u m' ( t )
 a3 y < a3
( um ( t ) )3 =  (4)
 η 3 – jθ y ≥ a3
 ------ e u m' ( t )
 y

where η 2 and η 3 are power normalization coefficients. The probability density function of
–y
power gain, y, in a Rayleigh channel with unity mean is p Y ( y ) = e . Hence, if the channel
is perfectly estimated and E m' denotes the symbol energy before compensation, then the
mean transmitted symbol energies are

( E m ) 1 = E m' (5)

∞ ∞
η2 e
–y
( E m ) 2 = ∫ ------ E m'e dy = η 2 E m' ∫ ------- dy
–y
(6)
y y
a2 a2

a3 ∞
η3 –y η3
( E m ) 3 = ∫ ------ e E m' dy + ∫ -----y- e
–y
E m' d y (7)
a3
0 a3
∞ –y
1 –a3 e 
= η 3 E m'  ----- ( 1 – e ) + ∫ ------- dy
 a3 a
y 
3
Paper III 3

In order to maintain the same transmitted mean energy for the three methods, the power nor-
malization coefficients are consequently chosen as

∞ ∞
 e–y  1 –a3 e
–y 
η 2 = 1 ⁄  ∫ ------- dy η 3 = 1 ⁄  ----- ( 1 – e ) + ∫ ------- dy (8)
a y   a3 a
y 
2 3

The instantaneous received symbol energy at the mobile terminal for the three compensation
methods becomes

( E m ) 1 = yE m' (9)

0 y < a2
( Em )2 =  (10)
 η 2 E m' y ≥ a2

y
 ----- η 3 E m' y < a3
( Em )3 =  a3 (11)
 η E ' y ≥ a3
 3 m

1
With QPSK the bit energy before compensation is E b' = --- E m' . The bit error rate, Pb, when
2
the Rayleigh fading channel is perfectly known is for the first method the same as for ordi-
nary coherent detection of QPSK signals, i.e. [3]

∞
 2E b' 1 E b' ⁄ N 0 
( Pb )1 = ∫ Q  y ---------- p Y ( y ) dy = ---  1 – --------------------------
N0  2 1 + E b' ⁄ N 0
- (12)
0

The bit error rate for the two other methods can be calculated as

a2 ∞
1  2E b'
( P b ) 2 = ∫ --- p Y ( y ) dy + ∫  ---------
Q  η 2 - p ( y ) dy (13)
2 N0  Y
0 a 2

1 –a2 –a2  2E b'

= --- ( 1 – e ) + e Q  η 2 ----------
2  N0 

a3 ∞
 η 3 2E b'  2E b'
( P b ) 3 = ∫ Q  y ------ ---------- p Y ( y ) dy + ∫ Q  η 3 ---------- p Y ( y ) dy (14)
0
 a 3 N 0 
a
 N0 
3

1 η 3 ( E b' ⁄ N 0 )  2E b' 1
- Q  2a 3 + η 3 ---------- – ---
= --- + ----------------------------------------
2 a 3 + η 3 ( E b' ⁄ N 0 )  N0  2
4 Paper III

The resulting BER is shown in Fig. 1 together with the curves for differential QPSK [3]. On
the left side of the figure too much energy is spent on compensating very bad channels,
while on the right side the transmitter is saturated or off even when communication normally
is advisable.

The compensation limits, a2 and a3, can be seen as a measure of the dynamic range for the
power compensation. However, in practical measurements the peak-to-average ratio of the
transmitted power is often used, which for the pre-compensation methods is the product of
the peak-to-average ratio of the modulation method and the increase due to the compensa-
tion. The latter is shown in Fig. 2, and can be calculated as 1, η 2 ⁄ a 2 and η 3 ⁄ a 3 for the three
methods respectively.

IV. Discussion
Phase-only compensation produces no increase in peak-to-average power ratio of the trans-
mitted signal, a favorable property for power amplifiers. The bit error rate is the same as
would be expected from a standard receiver with coherent demodulation, which is approxi-
mately half that obtained from differential detection. At low Eb’/N0 values phase-only cor-
rection appears to be best for the majority of a-values. At medium and high Eb’/N0 there is
some value in adding amplitude compensation with peak power saturation, particularly for
practical values of peak to average transmitter power increase (up to 10 dB). Switching off
the transmitter instead of saturating at the peaks to save power when the channel is poor,
appears to be only marginally viable at impracticably large peak to average power ratios, and
therefore should not be considered. The performance of this latter scheme is dominated by
the on/off duty cycle of the transmitter which explains the merging of the curves for high
values of the compensation limit, a. In comparison with differential decoding (where the ter-
minals have approximately the same complexity) all the compensation methods perform
well for appropriate compensation limits. The presented BER curves apply for both single-
and multicarrier systems (i.e. OFDM) and when both phase and amplitude are compensated
the technique can be extended to QAM systems as well.

V. Conclusions
BER when applying different methods of pre-compensating QPSK signals in a Rayleigh
fading channel is presented. It is shown that pre-compensation can be an attractive alterna-
tive to differential decoding, even when only phase is compensated and the amplitude is left
unaffected. If an additional 10 dB of signal variation is acceptable the bit error rate can be
reduced by an order of magnitude compared to traditional coherent detection and be reduced
by a factor of twenty compared to differential detection (Eb’/N0=20 dB).

References
[1] P. S. Henry and B. S. Glance, ”A new approach to high-capacity digital mobile
radio”, The Bell system technical journal, Vol. 60, No. 8, pp. 1891-1904, 1981.
[2] H. Matsuki and H. Takanashi, ”Theoretical analysis of BER improvement by
adaptive transmitter power control for Rayleigh fading compensation”, Electronics
Letters, 29, (17), pp. 1520-1521, 1993.
[3] J. G. Proakis, Digital communications, third edition, McGraw-Hill, New York,
USA, 1995.
Paper III 5

Figures

0
10 0
10
BER

-1
10 -1
10
Eb´/No=0 dB
-2
10 -2
-5 0 10
10 10
COMPENSATION LIMIT a

BER
0 -3
10 10

-1 -4
10 10
BER

-2 -5
10 10
Eb´/No=10 dB Eb´/No=20 dB
-3 -6
10 10
-5 0 -5 0
10 10 10 10
COMPENSATION LIMIT a COMPENSATION LIMIT a

Fig. 1. BER for the compensation methods when the channel is perfectly known,
Eb’/No= 0, 10 and 20 dB.
- - - differential
x method 1, phase compensation only.
+ method 2, Tx off when y<a2.
o method 3, Tx saturated when y<a3.

45
40
PEAK-TO-AVERAGE RATIO (dB)

35
30
25
20
15
10
5
0
-5
-4 -2 0
10 10 10
COMPENSATION LIMIT a

Fig. 2. Increase in peak to average power ratio for the three compensation
methods.
x method 1, phase compensation only.
+ method 2, Tx off when y<a2.
o method 3, Tx saturated when y<a3.
6 Paper III